As super-resolution (SR) techniques introduce unique distortions that fundamentally differ from those caused by traditional degradation processes (e.g., compression), there is an increasing demand for specialized video quality assessment (VQA) methods tailored to SR-generated content. One critical factor affecting perceived quality is temporal inconsistency, which refers to irregularities between consecutive frames. However, existing VQA approaches rarely quantify this phenomenon or explicitly investigate its relationship with human perception. Moreover, SR videos exhibit amplified inconsistency levels as a result of enhancement processes. In this paper, we propose Temporal Inconsistency Guidance for Super-resolution Video Quality Assessment (TIG-SVQA) that underscores the critical role of temporal inconsistency in guiding the quality assessment of SR videos. We first design a perception-oriented approach to quantify frame-wise temporal inconsistency. Based on this, we introduce the Inconsistency Highlighted Spatial Module, which localizes inconsistent regions at both coarse and fine scales. Inspired by the human visual system, we further develop an Inconsistency Guided Temporal Module that performs progressive temporal feature aggregation: (1) a consistency-aware fusion stage in which a visual memory capacity block adaptively determines the information load of each temporal segment based on inconsistency levels, and (2) an informative filtering stage for emphasizing quality-related features. Extensive experiments on both single-frame and multi-frame SR video scenarios demonstrate that our method significantly outperforms state-of-the-art VQA approaches.

This paper aims to interactively generate and edit disentangled 3D characters based on precise user instructions. Existing methods generate and edit 3D characters via rough and simple editing guidance and entangled representations, making it difficult to achieve precise and comprehensive control over fine-grained local editing and free clothing transfer for characters. To enable accurate and intuitive control over the generation and editing of high-quality 3D characters with freely interchangeable clothing, we propose a novel user-interactive approach for disentangled 3D character creation. Specifically, to achieve precise control over 3D character generation and editing, we introduce two user-friendly interaction approaches: a sketch-based layered character generation/editing method, which supports clothing transfer; and a 3D-proxy-based part-level editing method, enabling fine-grained disentangled editing. To enhance 3D character quality, we propose a 3D Gaussian reconstruction strategy guided by geometric priors, ensuring that 3D characters exhibit detailed local geometry and smooth global surfaces. Extensive experiments on both public datasets and in-the-wild data demonstrate that our approach not only generates high-quality disentangled 3D characters but also supports precise and fine-grained editing through user interaction.

Reference-based line art video colorization colorizes the target line art according to reference images, which is an essential stage for the cartoon production workflow. However, the manual colorization process is time-consuming and repetitive, making automatic video colorization highly desirable. Existing cartoon colorization methods struggle with domain misalignment between the reference and line art images and the loss of details caused by compression into a low-dimensional space in the existing video diffusion models, reducing colorization quality. In this paper, we propose an Offset Prior-based Diffusion Model (OPDM) for cartoon video colorization, which utilizes the powerful generation capability of the diffusion model and cross-domain matching priors to generate high-quality colorization results. Specifically, we design a simple and effective Offset-Adapter that leverages the idea of sampling offsets in deformable convolution to estimate the cross-domain spatial offset features between the target line arts and reference images. We further introduce a new training strategy that combines forward diffusion and reverse denoising in the training stage to ensure content consistency. Experiments on a public cartoon dataset and our newly constructed long cartoon video dataset demonstrate that our proposed method outperforms the existing state-of-the-art line art coloring methods. Code is available at https://github.com/xzh976/OPDM.

Exemplar-based image translation, which aims to transfer the style of an exemplar image to an input semantic image, is challenging and important in many applications. Most current methods build coarse correspondences and overlook extracting faithful style information from the exemplar image, leading to unsatisfactory results with style inconsistent with the exemplar image. In this paper, we propose a novel and efficient style mixture block to extract faithful style information and build reliable correspondences progressively. Specifically, instead of modeling explicit correspondences, we extract faithful style descriptors by considering global information about the exemplar features. Then, we generate coefficients for these style descriptors by modeling the interaction between the exemplar image and the input image, and efficiently compose these descriptors using the coefficients. The efficiency of the style mixture block allows a multi-scale architecture to extract and transform style descriptors at different resolutions, deforming the features of the exemplar image and refining the correspondences progressively. Experimental results on several datasets show that our SMixNet outperforms the current state-of-the-art, and is faster. Code is available for research purposes at https://github.com/Zhangjinso/SMixNet.

Existing dehazing methods primarily rely on synthetic hazy images for supervised learning. While effective on synthetic datasets, these methods often struggle to generalize to real-world hazy images, leading to issues such as color distortion and incomplete haze removal. Moreover, their limited adaptability to real-world datasets and inability to handle complex haze scenarios remain significant challenges. To address these limitations, we propose a novel unsupervised framework P3C-DNet (Pseudo-groundtruth Contrastive learning with Color Calibration Dehazing Network). Our P3C-DNet introduces a Pseudo-groundtruth image generation strategy through the Pseudo-groundtruth Contrastive Supervision (PCS) module, which overcomes the lack of real haze-free training data by generating high-quality Pseudo-groundtruth images. To further refine the dehazing process, we incorporate a codebook-based image coding and matching mechanism that aligns Pseudo-groundtruth images with hazy inputs, enhancing the accuracy and detail of the dehazed outputs. To address the prevalent issue of color distortion, especially in complex environments, our P3C-DNet integrates a Dynamic Color Restoration Block (DCRB) to ensure visual quality and color consistency in the dehazed results. Experimental evaluations demonstrate that our P3C-DNet achieves superior performance in haze removal, color fidelity, and detail preservation, significantly outperforming existing methods and setting a new benchmark for real-world dehazing tasks.

Due to the complex physical properties of granular materials, research on robot learning for manipulating such materials predominantly either disregards the consideration of their physical characteristics or uses surrogate models to approximate their physical properties. Learning to manipulate granular materials based on physical information obtained through precise modeling remains an unsolved problem. In this article, we propose to address this challenge by constructing a differentiable physics-based simulator for granular materials using the Taichi programming language and developing a learning framework accelerated by demonstrations generated through gradient-based optimization on nongranular materials within our simulator, eliminating the costly data collection and model training of prior methods. Experimental results show that our method, with its flexible design, trains robust policies that are capable of executing the task of transporting granular materials in both simulated and real-world environments, beyond the capabilities of standard reinforcement learning (RL), imitation learning (IL), and prior task-specific granular manipulation methods.

Automating the manipulation of granular materials poses significant challenges due to complex contact dynamics, unpredictable material properties, and intricate system states. Existing approaches often fail to achieve efficiency and accuracy in such tasks. To fill the research gap, this paper studies the small-scale and high-precision granular material digging task with unknown physical properties. A key scientific problem addressed is the feasibility of applying first-order gradient based optimisation to complex differentiable granular material simulation and overcoming associated numerical instability. A new framework, named differentiable digging robot (DDBot), is proposed to manipulate granular materials, including sand and soil. Specifically, we equip DDBot with a differentiable physics based simulator, tailored for granular material manipulation, powered by GPU-accelerated parallel computing and automatic differentiation. DDBot can perform efficient differentiable system identification and high-precision digging skill optimisation for unknown granular materials, which is enabled by a differentiable skill-to-action mapping, a task-oriented demonstration method, gradient clipping and line search-based gradient descent. Experimental results show that DDBot can efficiently (converge within 5 to 20 minutes) identify unknown granular material dynamics and optimise digging skills, with high-precision results in zero-shot real-world deployments, highlighting its practicality. Benchmark results against state-of-the-art baselines also confirm the robustness and efficiency of DDBot in such digging tasks.

Welsh is a linguistically rich yet under-resourced minority language. Despite its cultural significance, automated fluency assessment remains largely unexplored due to limited datasets and tools. Existing models focus on high-resource languages, leaving Welsh without sufficient multi-modal resources. To address this, we introduce CymruFluency, the first 4D dataset for Welsh fluency assessment, capturing both audio and 3D lip movements with expert-annotated fluency scores. Building on this, we propose a multi-modal fluency classification framework that combines audio features (mel spectrograms) and manually annotated 3D lip landmarks. Our fusion approach significantly improves fluency prediction over unimodal models, emphasizing the critical role of 3D lip dynamics in Welsh learning. This research advances minority language processing by integrating articulatory features into fluency evaluation, offering a powerful tool for Welsh language learning, assessment, and preservation. Project page: https://github.com/arvinsingh/CymruFluency .

Generating and creating humanoid 3D models has received increasing attention recently due to its fundamental support for many high-level 3D applications. Although automatic 3D pose and shape reconstruction methods have achieved promising results, there are still some failure cases due to self-occlusions, viewpoint changes, and the complexity of human pose articulations. In this paper, we propose a novel way to leverage Large Language Models (LLMs) to interactively reconstruct human pose and shape based on a Skinned Multi-Person Linear (SMPL) model. We construct a mapping table to fine-tune an LLM, enabling it to understand user inputs better and output the positional information of joint points. Additionally, a simple neural network is adopted to regress the shape cues of the SMPL. We demonstrate a gallery of results of numerous poses and shapes. We validate our method via numerical evaluations, user studies, and comparisons to manually posed characters and previous work.

Multi-class part parsing is a dense prediction task that segments objects into semantic components with multi-level abstractions. Despite its significance, this task remains challenging due to ambiguities at both part and class levels. In this paper, we propose a network that incorporates multi-class boundaries to precisely identify and emphasize the spatial boundaries of part classes, thereby improving segmentation quality. Additionally, we employ a weighted multi-label cross-entropy loss function to ensure balanced and effective learning from all parts. Experimental results validate the effectiveness of the proposed method, demonstrating its ability to enhance baseline performance on benchmark datasets.

Audio-driven talking face generation from a single source image is a popular research topic. There still exist many challenges for its practical applications, e.g., diverse motion generation, effective emotional control, and large view angle changes. In this work, we propose a novel one-shot emotion-controllable audio-driven 3D talking face generation framework, which creates free-view talking videos from one reference image. Firstly, to synchronize the motion with the input audio, we use a transformer-based motion generator to capture the context of the input audio and predict motion coefficient sequences, which are leveraged by a motion encoder to extract motion codes. Meanwhile, to reconstruct a 3D portrait from one reference image, an identity encoder is utilized to extract an identity code and generate emotion-dependent appearance with a specific emotion label. Finally, we introduce an emotion-controllable 3D portrait video generator to synthesize free-view talking videos using the disentangled motion and identity codes. Thanks to the audio-synchronized motion codes and emotion-aware identity code, we can render a talking face with realistic emotional expressions in novel views. Extensive experiments show that our method is capable of maintaining superior visual performance and motion accuracy in both front view and novel views.

Robotic soil manipulation is essential for automated farming, particularly in excavation and levelling tasks. However, the nonlinear dynamics of granular materials challenge traditional control methods, limiting stability and efficiency. We propose Celebi, a causality-enhanced optimisation method that integrates differentiable physics simulation with adaptive step-size adjustments based on causal inference. To enable gradient-based optimisation, we construct a differentiable simulation environment for granular material interactions. We further define skill parameters with a differentiable mapping to end-effector motions, facilitating efficient trajectory optimisation. By modelling causal effects between task-relevant features extracted from point cloud observations and skill parameters, Celebi selectively adjusts update step sizes to enhance optimisation stability and convergence efficiency. Experiments in both simulated and real-world environments validate Celebi’s effectiveness, demonstrating robust and reliable performance in robotic excavation and levelling tasks.

Rolling contact kinematics plays a vital role in dexterous manipulation and rolling-based locomotion. Yet, in practical applications, the environments and objects involved are often captured as discrete point clouds, creating substantial difficulties for traditional motion control and planning frameworks that rely on continuous surface representations. In this work, we propose a differential geometry-based framework that models point cloud data for continuous rolling contact using locally parameterized representations. Our approach leverages skeletonization to define a rotational reference structure for rolling interactions and applies a Fourier-based curve fitting technique to extract and represent meaningful controllable local geometric structure. We further introduce a novel 2D manifold coordinate system tailored to arbitrary surface curves, enabling local parameterization of complex shapes. The governing kinematic equations for rolling contact are then derived, and we demonstrate the effectiveness of our method through simulations on various object examples.

Backchannel signals play a critical role in social interaction, expressing attentiveness, agreement, and emotion in both human and human-agent conversations. However, few multi-modal databases exist in this area due to the complexity of categorisation and the high cost of precise timing, especially in naturalistic dyadic conversations. To address these challenges, we introduce CCDb+ (Cardiff Conversation Database +) an enhanced version of CCDb, with 25 newly annotated conversations and corrections to 14 previously annotated conversations, along with thorough consistency checks to ensure annotation reliability. Additionally, we propose a multi-modal process for backchannel detection as a baseline, showing that both visual and acoustic cues contribute significantly to understanding backchannel behaviour. Recognising that backchannel signals often intersect with other social cues, we introduce several detection sub-tasks-such as smile, nodding, and agreement-with baseline results for each. Finally, we demonstrate multi-modal paradigms for nuanced signals like nodding and thinking. The database and associated annotations are publicly available at https://huggingface.co/datasets/CardiffVisualComputing/CCDb.

Creating high-quality, photorealistic 3D digital humans from a single image remains challenging. While existing methods can generate visually appealing multi-view outputs, they often suffer from inconsistencies in viewpoints and camera poses, resulting in suboptimal 3D reconstructions with reduced realism. Furthermore, most approaches focus on body generation while overlooking facial consistency – a perceptually critical issue caused by the fact that the face occupies only a small area in a full-body image (e.g., ∼ 80 × 80 pixels out of a 512 × 512 image). This limited resolution and low weight for the facial regions during optimization leads to insufficient facial details and inconsistent facial identity features across multiple views.To address these challenges, we leverage the powerful capabilities of 2D video diffusion models for consistent multi-view RGB and Normal human image generation, combined with the 3D SMPL-X representation to enable spatial consistency and geometrical details. By fine-tuning the DiT models (HumanWan-DiTs) on realistic 3D human datasets using the LoRA technique, our method ensuresboth generalizability and 3D visual consistency on realistic multi-view human image generation. The proposed facial enhancement is integrated into 3D Gaussian optimization to enhance facial details. To further refine results, we apply super-resolution and generative priors to reduce facial blurring alongside SMPL-X parameter tuning and the assistance of generated multi-view normal images, achieving photorealistic and consistent rendering from a single image. Extensive experiments demonstrate that our approach outperforms existing methods, producing photorealistic, consistent, and fine-detailed human renderings.

Mirror detection aims to identify mirror areas in a scene, with recent methods either integrating depth information (RGB-D) or making use of temporal information (video). However, utilizing both data is still under-explored due to the lack of a high-quality dataset and an effective method for the RGB-D Video Mirror Detection (DVMD) problem. To the best of our knowledge, this is the first work to address the DVMD problem. To exploit depth and temporal information in mirror segmentation, we first construct a large-scale RGB-D Video Mirror Detection Dataset (DVMD-D), which contains 17977 RGB-D images from 273 diverse videos. We further develop a novel model, named DVMDNet, which can first locate the mirrors based on triple consistencies: local consistency, cross-modality consistency and global consistency, and then refine the mirror boundaries through content discontinuity, taking the temporal information within videos into account. We conduct a comparative study on the DVMD dataset, evaluating 12 state-of-the-art models (including single-image mirror detection, single-image glass detection, RGB-D mirror detection, video shadow detection, video glass detection, and video mirror detection methods). Code is available from https://github.com/UpChen/2025_DVMDNet.

Novel space-time view synthesis for monocular video is a highly challenging task: both static and dynamic objects usually appear in the video, but only a single view of the current scene is available, resulting in inaccurate synthesis results. To address this challenge, we propose FRNeRF, a novel space-time view synthesis method with a fusion regularization field. Specifically, we design a 2D-3D fusion regularization field for the original dynamic neural field, which helps reduce blurring of dynamic objects in the scene. In addition, we add image prior features to the hierarchical sampling to solve the problem that the traditional hierarchical sampling strategy cannot obtain sufficient sampling points during training. We evaluate our method extensively on multiple datasets and show the results of dynamic space-time view synthesis. Our method achieves state-of-the-art performance both qualitatively and quantitatively. Code is available for research purposes at https://cic.tju.edu.cn/faculty/likun/projects/FRNerf.

Video instance segmentation (VIS) is an evolving research topic in computer vision that aims to simultaneously detect, segment, and track semantic objects across multiple video frames. However, existing VIS methods are typically unaware of the reliability of the training samples from insufficient and imbalanced datasets, leading to suboptimal performance. To address this challenge, we propose a memory-based conditional neural process (MemCNP) module to exploit the strengths of both memory networks and the CNP model which handles heterogeneous latent space distributions for reliable modelling with insufficient data. Our MemCNP utilises predicted uncertainty to regularise VIS predictions as well as to identify reliable samples for effective training. Notably, our MemCNP is model-agnostic and can thus be seamlessly integrated into various VIS models to improve their performance. Extensive experiments on the YouTube-VIS and OVIS datasets demonstrate the effectiveness of MemCNP regardless of the underlying model architecture. • A memory-based conditional neural process. • Reliability modelling for object detection. • Uncertainty-based dynamic training sample selection. • Contrastive instance tracking.

Neural Radiance Fields (NeRFs) have shown promising results in novel view synthesis. While achieving state-of-the-art rendering results, NeRF usually encodes all properties related to geometry and appearance of the scene together into several MLP (Multi-Layer Perceptron) networks, which hinders downstream manipulation of geometry, appearance and illumination. Recently researchers made attempts to edit geometry, appearance and lighting for NeRF. However, they fail to render view-consistent results after editing the appearance of the input scene. Moreover, many approaches use Spherical Gaussian (SG) or Spherical Harmonic (SH) functions, or low-resolution environment maps to model lighting. These methods, however, struggle with high-frequency environmental relighting. While some approaches utilize high-resolution environment maps, the strategy of jointly optimizing geometry, material, and lighting introduces additional ambiguity. To solve the above problems, we propose VD-NeRF, a visibility-aware approach to decoupling view-independent appearance and view-dependent appearance in the scene with a hybrid lighting representation. Specifically, we first train a signed distance function to reconstruct an explicit mesh for the input scene. Then a decoupled NeRF learns to attach view-independent appearance to the reconstructed mesh by defining learnable disentangled features representing geometry and view-independent appearance on its vertices. For lighting, we approximate it with an explicit learnable environment map and an implicit lighting network to support both low-frequency and high-frequency relighting. By modifying the view-independent appearance, rendered results are consistent across different viewpoints. Our method also supports high-frequency environmental relighting by replacing the explicit environment map with a novel one and fitting the implicit lighting network to the novel environment map. We further take visibility into consideration when rendering and decoupling the input 3D scene, which improves the quality of decomposition and relighting results and also enables more downstream applications such as scene composition where occlusions between scenes are common. Extensive experiments show that our method achieves better editing and relighting performance both quantitatively and qualitatively compared to previous methods.

Reconstructing and editing 3D objects and scenes both play crucial roles in computer graphics and computer vision. Neural radiance fields (NeRFs) can achieve realistic reconstruction and editing results but suffer from inefficiency in rendering. Gaussian splatting significantly accelerates rendering by rasterizing Gaussian ellipsoids. However, Gaussian splatting utilizes a single Spherical Harmonic (SH) function to model both texture and lighting, limiting independent editing capabilities of these components. Recently, attempts have been made to decouple texture and lighting with the Gaussian splatting representation but may fail to produce plausible geometry and decomposition results on reflective scenes. Additionally, the forward shading technique they employ introduces noticeable blending artifacts during relighting, as the geometry attributes of Gaussians are optimized under the original illumination and may not be suitable for novel lighting conditions. To address these issues, we introduce DeferredGS, a method for decoupling and relighting the Gaussian splatting representation using deferred shading. To achieve successful decoupling, we model the illumination with a learnable environment map and define additional attributes such as texture parameters and normal direction on Gaussians, where the normal is distilled from a jointly trained signed distance function. More importantly, we apply deferred shading, resulting in more realistic relighting effects compared to previous methods. Both qualitative and quantitative experiments demonstrate the superior performance of DeferredGSin novel view synthesis and relighting tasks.

In the challenging realm of image-to-image translation, most traditional methods require separate models for different translation directions, leading to inefficient use of computational resources. This paper introduces the Bidirectional Brownian Bridge Diffusion Model (BiBBDM), a novel approach that leverages Brownian Bridge processes for bidirectional image-to-image translation. Unlike conventional Diffusion Models (DMs) that treat image-to-image translation as a unidirectional conditional generation process, BiBBDM models the translation as a stochastic Brownian Bridge process, enabling simultaneous learning of bidirectional translation between two domains. This innovation allows our method to achieve bidirectional image translation using different sampling directions of a single model, eliminating the need for multiple models for both translation directions. To the best of our knowledge, BiBBDM is the first image translation framework to achieve simultaneous dual-domain sampling with the same model and parameters, based on Brownian Bridge diffusion processes. Extensive experimental results on various benchmarks demonstrate that BiBBDM achieves competitive performance, as evidenced by both visual inspection and quantitative metrics.

Object Goal Navigation (ObjectNav) refers to an agent navigating to an object in an unseen environment, which is an ability often required in the accomplishment of complex tasks. Though it has drawn increasing attention from researchers in the Embodied AI community, there has not been a contemporary and comprehensive survey of ObjectNav. In this survey, we give an overview of this field by summarizing more than 70 recent papers. First, we give the preliminaries of the ObjectNav: the definition, the simulator, and the metrics. Then, we group the existing works into three categories: 1) end-to-end methods that directly map the observations to actions, 2) modular methods that consist of a mapping module, a policy module, and a path planning module, and 3) zero-shot methods that use zero-shot learning to do navigation. Finally, we summarize the performance of existing works and the main failure modes and discuss the challenges of ObjectNav. This survey would provide comprehensive information for researchers in this field to have a better understanding of ObjectNav.Note to Practitioners—This work was motivated by the increased interest in real-world applications of mobile robots. Object Goal Navigation (ObjectNav), which is an important task in these applications, requires an agent to find an object in an unseen environment. To accomplish that, the agent needs to be equipped with the capability to move in the environment, decide where to go, and recognize the object categories. So far, most works on ObjectNav have been done in a simulation environment. We present an overview of the existing works in ObjectNav and introduce them in three categories. Additionally, we analyze the current performance of ObjectNav and the challenges for future research. This paper provides researchers and practitioners with a comprehensive overview of the developed methods in ObjectNav, which can help them to have a good understanding of this task and develop suitable solutions for applications in the real world.

Most collection-based style transfer methods require training a separate model for each individual collection of styles, making the extension to multiple collections of styles less flexible. Besides, the existing collection-based methods are also less flexible in extending to new style collections in a continual manner. To address these issues, we propose a novelMultI-Dictionary Generative Adversarial Network framework (MID-GAN)for multi-collection style transfer. Specifically, we design a multi-dictionary architecture within a GAN, with each dictionary consisting of a set of local style codes for a specific style collection. Benefiting from the local style codes used in the dictionary, a stylization module with aligned skip connections is further proposed, which can better preserve both the local details and the overall image structure. The dictionary design allows a flexible extension to new style collections by readily adding new dictionaries and we propose a continual training strategy that can both preserve the style transfer ability of old styles and achieve good transfer results for newly added styles. Extensive experiments are performed to show that the proposed method is better than existing collection-based style transfer methods. We also demonstrate the proposed method can generate diverse meaningful style transfer results of the same style collection.

Animatable and relightable 3D facial generation has fundamental applications in computer vision and graphics. Although animation and relighting are highly correlated, previous methods usually address them separately. Effectively combining animation methods and relighting methods is nontrivial. In terms of explicit shading models, animatable methods cannot be easily extended to achieve realistic relighting results, such as shadow effects, due to prohibitive computational training costs. Regarding implicit lighting representations, current animatable methods cannot be incorporated due to their inharmonious animation representations, i.e., deforming spatial points. This paper, armed with a lightweight but effective lighting representation, presents a compatible animation representation to achieve a disentangled generative model of 3D animatable and relightable heads. Our represented animation allows for updating and control of realistic lighting effects. Due to the disentangled nature of our representations, we learn the animation and relighting from large-scale, in-the-wild videos instead of relying on a morphable model. We show that our method can synthesize geometrically consistent and detailed motion along with the disentangled control of lighting conditions. We further show that our method is still compatible with morphable models for driving generated avatars. Our method can also be extended to domains without video data by domain transfer to achieve a broader range of animatable and relightable head synthesis. We will release the code for reproducibility and facilitating future research.

Most of the existing 3D talking face synthesis methods suffer from the lack of detailed facial expressions and realistic head poses, resulting in unsatisfactory experiences for users. In this article, we propose a novel pose-aware 3D talking face synthesis method with a novel geometry-guided audio-vertices attention. To capture more detailed expression, such as the subtle nuances of mouth shape and eye movement, we propose to build hierarchical audio features including a global attribute feature and a series of vertex-wise local latent movement features. Then, in order to fully exploit the topology of facial models, we further propose a novel geometry-guided audio-vertices attention module to predict the displacement of each vertex by using vertex connectivity relations to take full advantage of the corresponding hierarchical audio features. Finally, to accomplish pose-aware animation, we expand the existing database with an additional pose attribute, and a novel pose estimation module is proposed by paying attention to the whole head model. Numerical experiments demonstrate the effectiveness of the proposed method on realistic expression and head movements against state-of-the-art methods.

Stroke-based Rendering (SBR) aims to decompose an input image into a sequence of parameterized strokes, which can be rendered into a painting that resembles the input image. Recently, Neural Painting methods that utilize deep learning and reinforcement learning models to predict the stroke sequences have been developed, but suffer from longer inference time or unstable training. To address these issues, we propose AttentionPainter, an efficient and adaptive model for single-step neural painting. First, we propose a novel scalable stroke predictor, which predicts a large number of stroke parameters within a single forward process, instead of the iterative prediction of previous Reinforcement Learning or auto-regressive methods, which makes AttentionPainter faster than previous neural painting methods. To further increase the training efficiency, we propose a Fast Stroke Stacking algorithm, which brings 13 times acceleration for training. Moreover, we propose Stroke-density Loss, which encourages the model to use small strokes for detailed information, to help improve the reconstruction quality. Finally, we design a Stroke Diffusion Model as an application of AttentionPainter, which conducts the denoising process in the stroke parameter space and facilitates stroke-based inpainting and editing applications helpful for human artists' design. Extensive experiments show that AttentionPainter outperforms the state-of-the-art neural painting methods.

Radiance field based methods have recently revolutionized novel view synthesis of scenes captured with multi-view photos. A significant recent advance is 3D Gaussian Splatting (3DGS), which utilizes a set of 3D Gaussians to represent a radiance field, yielding high-fidelity results in real-time rendering. However, we have observed that 3DGS struggles to capture the necessary details in sparsely observed regions, where there is not enough gradient for effective split and clone operations. In this paper, we present a novel solution to address this limitation. Our key idea is to leverage segmentation information to identify poorly optimized regions within the 3D Gaussian representation. By applying split or clone operations on the corresponding 3D Gaussians in these regions, we aim to refine the spatial distribution of Gaussians and enhance the overall quality of high-fidelity 3D scene reconstruction. To further optimize the reconstruction process, we introduce two spatial regularization terms: repulsion loss and smoothness loss. These terms effectively minimize overlap and redundancy among Gaussians, reducing outliers in the synthesized geometry. By incorporating these regularization techniques, our approach achieves state-of-the-art performance in real-time novel view synthesis and significantly improves visibility in less observed regions, leading to a more compact and accurate 3D scene representation.

Most people with a visual impairment retain some residual vision, yet accessibility strategies and research for visual art focus on non-visual aids and strategies. Little is known about how image enhancements may benefit people with low vision in art settings. Our study explored the challenges of people with visual impairments in museums and art galleries and the potential of visual enhancements to improve accessibility. Through online focus groups with 18 participants, we uncovered multifaceted visual challenges. Participants emphasised the role of non-visual senses to reduce visual ambiguities. Subsequently, we presented participants with the concept of image enhancements on Head-Mounted Displays (HMDs) as a potential approach to enhance visual accessibility. Participants expressed nuanced perspectives, advocating for this technology to improve visual art. The findings provide valuable insights for enhancing the visual experience of people with low vision in museums and galleries through design.

Online handwriting recognition is pivotal in domains like note-taking, education, healthcare, and office tasks. Existing diagram recognition algorithms mainly rely on the temporal information of strokes, resulting in a decline in recognition performance when dealing with notes that have been modified or have no temporal information. The current datasets are drawn based on templates and cannot reflect the real free-drawing situation. To address these challenges, we present SpaceGTN, a time-agnostic Graph Transformer Network, leveraging spatial integration and removing the need for temporal data. Extensive experiments on multiple datasets have demonstrated that our method consistently outperforms existing methods and achieves state-of-the-art performance. We also propose a pipeline that seamlessly connects offline and online handwritten diagrams. By integrating a stroke restoration technique with SpaceGTN, it enables intelligent editing of previously uneditable offline diagrams at the stroke level. In addition, we have also launched the first online handwritten diagram dataset, OHSD, which is collected using a free-drawing method and comes with modification annotations.

Synthesizing realistic videos of talking faces under custom lighting conditions and viewing angles benefits various downstream applications like video conferencing. However, most existing relighting methods are either time-consuming or unable to adjust the viewpoints. In this paper, we present the first real-time 3D-aware method for relighting in-the-wild videos of talking faces based on Neural Radiance Fields (NeRF). Given an input portrait video, our method can synthesize talking faces under both novel views and novel lighting conditions with a photo-realistic and disentangled 3D representation. Specifically, we infer an albedo tri-plane, as well as a shading tri-plane based on a desired lighting condition for each video frame with fast dual-encoders. We also leverage a temporal consistency network to ensure smooth transitions and reduce flickering artifacts. Our method runs at 32.98 fps on consumer-level hardware and achieves state-of-the-art results in terms of reconstruction quality, lighting error, lighting instability, temporal consistency and inference speed. We demonstrate the effectiveness and interactivity of our method on various portrait videos with diverse lighting and viewing conditions.

SVG (Scalable Vector Graphics) is a widely used graphics format that possesses excellent scalability and editability. Image vectorization, which aims to convert raster images to SVGs, is an important yet challenging problem in computer vision and graphics. Existing image vectorization methods either suffer from low reconstruction accuracy for complex images or require long computation time. To address this issue, we propose SuperSVG, a superpixel-based vectorization model that achieves fast and high-precision image vectorization. Specifically, we decompose the input image into superpixels to help the model focus on areas with similar colors and textures. Then, we propose a two-stage self-training framework, where a coarse-stage model is employed to reconstruct the main structure and a refinement-stage model is used for enriching the details. Moreover, we propose a novel dynamic path warping loss to help the refinement-stage model to inherit knowledge from the coarse-stage model. Extensive qualitative and quantitative experiments demonstrate the superior performance of our method in terms of reconstruction accuracy and inference time compared to state-of-the-art approaches. The code is available in https://github.com/sjtuplayer/SuperSVG.

Estimating 3D hand pose and recovering the full hand surface mesh from a single RGB image is a challenging task due to self-occlusions, viewpoint changes, and the complexity of hand articulations. In this paper, we propose a novel framework that combines an attention mechanism with heatmap regression to accurately and efficiently predict 3D joint locations and reconstruct the hand mesh. We adopt a pooling attention module that learns to focus on relevant regions in the input image to extract better features for handling occlusions, while greatly reducing the computational cost. The multi-scale 2D heatmaps provide spatial constraints to guide the 3D vertex predictions. By exploiting the complementary strengths of sparse 2D supervision and dense mesh regression, our method accurately reconstructs hand meshes with realistic details. Extensive experiments on standard benchmarks demonstrate that the proposed method efficiently improves the performance of 3D hand pose estimation and mesh recovery. The reproducible recipes are available at https://github.com/SDiannn/AHRNET-Heatmap.

Vector graphics are widely used in graphical designs and have received more and more attention. However, unlike raster images which can be easily obtained, acquiring high-quality vector graphics, typically through automatically converting from raster images, remains a significant challenge, especially for more complex images such as photos or artworks. In this paper, we propose SAMVG, a multi-stage model to vectorize raster images into SVG (Scalable Vector Graphics). Firstly, SAMVG uses general image segmentation provided by the Segment-Anything Model and uses a novel filtering method to identify the best dense segmentation map for the entire image. Secondly, SAMVG then identifies missing components and adds more detailed components to the SVG. Through a series of extensive experiments, we demonstrate that SAMVG can produce high quality SVGs in any domain while requiring less computation time and complexity compared to previous state-of-the-art methods.

Multi-class part parsing is a dense prediction task that decomposes objects into semantic components with multi-level abstractions. Despite the importance of this problem, it remains challenging due to the presence of both part-level and class-level ambiguities. In this paper, we propose GRPSNet network which integrates graph reasoning to capture relationships between parts for part segmentation. These captured relationships help to enhance the recognition and localization of parts. We also propose to exploit the relationships of part boundaries to further enhance the accuracy of part segmentation. The experimental results demonstrate the effectiveness of the proposed method and show that it achieves state-of-the-art performance on the benchmark datasets.

Techniques for detecting mirrors from static images have witnessed rapid growth in recent years. However, these methods detect mirrors from single input images. Detecting mirrors from video requires further consideration of temporal consistency between frames. We observe that humans can recognize mirror candidates, from just one or two frames, based on their appearance (e.g. shape, color). However, to ensure that the candidate is indeed a mirror (not a picture or a window), we often need to observe more frames for a global view. This observation motivates us to detect mirrors by fusing appearance features extracted from a short-term attention module and context information extracted from a long-term attention module. To evaluate the performance, we build a challenging benchmark dataset of 19,255 frames from 281 videos. Experimental results demonstrate that our method achieves state-of-the-art performance on the benchmark dataset.

Despite impressive results, deep generative models require massive datasets for training, and as dataset size increases, effective evaluation metrics like precision and recall (P&R) become computationally infeasible on commodity hardware. In this paper, we address this challenge by proposing efficient P&R (eP&R) metrics that give almost identical results as the original P&R but with much lower computational costs. Specifically, we identify two redundancies in the original P&R: i) redundancy in ratio computation and ii) redundancy in manifold inside/outside identification. We find both can be effectively removed via hubness-aware sampling, which extracts representative elements from synthetic/real image samples based on their hubness values, i.e., the number of times a sample becomes a k-nearest neighbor to others in the feature space. Thanks to the insensitivity of hubness-aware sampling to exact k-nearest neighbor (k-NN) results, we further improve the efficiency of our eP&R metrics by using approximate k-NN methods. Extensive experiments show that our eP&R matches the original P&R but is far more efficient in time and space. Our code is available at: https://github.com/Byronliang8/Hubness_Precision_Recall

Content-based 360° video editing allows users to manipulate panoramic content for interaction in a dynamic visual world. However, the current related methods (2D neural representation and optical flow) show limitations in producing high-quality panoramic content from 360° videos due to their lack of capacity to model the inherent spatiotemporal relationships among pixels in the true panoramic space. To address this issue, we propose a Neural Panoramic Representation (NPR) method to model the global inter-pixel relationships, facilitating immersive video editing. Specifically, our method utilizes MLP-based networks to learn spherical implicit content layers, by encoding the spherical spatiotemporal positions and appearance details within the panoramic video, and bi-directional mapping between the original video frames and the learned content layers, to capture the interpretable and global omnidirectional visual characteristics of individual dynamic scenes. Additionally, we introduce innovative loss functions (spherical neighborhood consistency and unit spherical regularization) to ensure the creation of appropriate implicit spherical content layers. We further provide an interactive layer neural panoramic editing approach based on the proposed NPR, in the head-mounted display device. We evaluate this framework on diverse real-world 360° videos, showing superior performance on both reconstruction and consistent editing compared to existing state-of-the-art (SOTA) neural representation techniques.

This paper aims to generate physically-layered 3D humans from text prompts. Existing methods either generate 3D clothed humans as a whole or support only tight and simple clothing generation, which limits their applications to virtual try-on and partlevel editing. To achieve physically-layered 3D human generation with reusable and complex clothing, we propose a novel layer-wise dressed human representation based on a physically-decoupled diffusion model. Specifically, to achieve layer-wise clothing generation, we propose a dual-representation decoupling framework for generating clothing decoupled from the human body, in conjunction with an innovative multi-layer fusion volume rendering method. To match the clothing with different body shapes, we propose an SMPL-driven implicit field deformation network that enables the free transfer and reuse of clothing. Extensive experiments demonstrate that our approach not only achieves state-of-the-art layered 3D human generation with complex clothing but also supports virtual try-on and layered human animation. More results and the code can be found on our project page at https: //cic.tju.edu.cn/faculty/likun/projects/HumanCoser

Rendering 3D human appearance from a single image in real-time is crucial for achieving holographic communication and immersive VR/AR. Existing methods either rely on multi-camera setups or are constrained to offline operations. In this paper, we propose R2Human, the first approach for real-time inference and rendering of photorealistic 3D human appearance from a single image. The core of our approach is to combine the strengths of implicit texture fields and explicit neural rendering with our novel representation, namely Z-map. Based on this, we present an end-to-end network that performs high-fidelity color reconstruction of visible areas and provides reliable color inference for occluded regions. To further enhance the 3D perception ability of our network, we leverage the Fourier occupancy field as a prior for generating the texture field and providing a sampling surface in the rendering stage. We also propose a consistency loss and a spatial fusion strategy to ensure the multi-view coherence. Experimental results show that our method outperforms the state-of-the-art methods on both synthetic data and challenging real-world images, in real-time. The project page can be found at http://cic.tju. edu.cn/faculty/likun/projects/R2Human.

Recent studies have shown impressive progress in universal style transfer which can integrate arbitrary styles into content images. However, existing approaches struggle with low aesthetics and disharmonious patterns in the final results. To address this problem, we propose AesStyler, a novel Aesthetic Guided Universal Style Transfer method. Specifically, our approach introduces the aesthetic assessment model, trained on a dataset with human-assessed aesthetic scores, into the universal style transfer task to accurately capture aesthetic features that universally resonate with human aesthetic preferences. Unlike previous methods which only consider aesthetics of specific style images, we propose to build a Universal Aesthetic Codebook (UAC) to harness universal aesthetic features that encapsulate the global aspects of aesthetics. Aesthetic features are fed into a novel Universal and Style-specific Aesthetic-Guided Attention (USAesA) module to guide the style transfer process. USAesA empowers our model to integrate the aesthetic attributes of both universal and style-specific aesthetic features with style features and facilitates the fusion of these aesthetically enhanced style features with content features. Extensive experiments and user studies have demonstrated that our approach generates aesthetically more harmonious and pleasing results than the state-ofthe- art methods, both aesthetic-free and aesthetic-aware. The code is available at: https://github.com/zwandering/AesStyler.

Scene synthesis has gained significant attention recently, and interactive scene synthesis focuses on yielding scenes according to user preferences. Existing literature either generates floor plans or scenes according to the floor plans. The system proposed in this paper generates scenes over floor plans in real-time. Given an initial scene, the only interaction a user needs is changing the room shapes. Our framework splits/merges rooms and adds/rearranges/removes objects for each transient moment during interactions. A systematic pipeline achieves our framework by compressing objects' arrangements over modified room shapes in a transient moment, thus enabling real-time performances. We also propose elastic boxes that indicate how objects should be arranged according to their continuously changed contexts, such as room shapes and other objects. Through a few interactions, a floor plan filled with object layouts is generated concerning user preferences on floor plans and object layouts according to floor plans. Experiments show that our framework is efficient at user interactions and plausible for synthesizing 3D scenes.

In semantic navigation, a top-down map with accurate and complete semantic information is vital to subsequent decision-making. However, due to occlusions and limitations of the robot’s field of view (FOV), there are often unobserved areas in the top-down maps. To address this problem, recent works have studied semantic map prediction to complete the top-down maps. In this work, we propose to improve map prediction by integrating relational information. We propose RSMPNet, a relationship-guided semantic map prediction network, which makes use of semantic and spatial relationships to predict unobserved areas from accumulated semantic maps. Specifically, we propose a Relationship Reasoning Layer that includes two modules, namely 1) the Semantic Relationship Graph Reasoning Module (SeGRM) to capture the semantic relationship and 2) the Spatial Relationship Graph Reasoning Module (SpGRM) to utilize the spatial relationship. We also design a semantic relationship enhanced loss to enhance our model to learn semantic relationship information. Experiments show the effectiveness of our proposed network which achieves state-of-the-art performance in semantic map prediction. Our code and dataset are publicly available at https://github.com/jws39/semantic-map-prediction

Reconstructing accurate 3D surfaces from noisy point clouds is a fundamental problem in computer vision. Among different approaches, neural implicit methods that map 3D coordinates to occupancy values benefit from the learning capabilities of deep neural networks and the flexible topology of implicit representations, achieving promising reconstruction results. However, existing methods utilize standard (dense) 3D convolutional neural networks for feature extraction and occupancy prediction, which significantly restricts their capability to reconstruct details. In this paper, we propose a neural implicit method based on sparse convolutions, where features and network calculations only focus on grid points close to the surface to be reconstructed. This allows us to build significantly higher resolution 3D grids and reconstruct high-fidelity details. We further build a 3D residual UNet to extract features which are robust to noise, while ensuring details are retained. A 3D position along with features extracted at the position are fed into the occupancy probability predictor network to obtain occupancy. As features at nearby grid points to the query position may not exist due to the sparse nature, we propose a normalized weight interpolation approach to obtain smooth interpolation with sparse data. Experimental results demonstrate that our method achieves promising results, both qualitatively and quantitatively, outperforming existing methods.

The cross-view matching of local image features is a fundamental task in visual localization and 3D reconstruction. This study proposes FilterGNN, a transformer-based graph neural network (GNN), aiming to improve the matching efficiency and accuracy of visual descriptors. Based on high matching sparseness and coarse-to-fine covisible area detection, FilterGNN utilizes cascaded optimal graph-matching filter modules to dynamically reject outlier matches. Moreover, we successfully adapted linear attention in FilterGNN with post-instance normalization support, which significantly reduces the complexity of complete graph learning from O ( N 2 ) to O ( N ). Experiments show that FilterGNN requires only 6% of the time cost and 33.3% of the memory cost compared with SuperGlue under a large-scale input size and achieves a competitive performance in various tasks, such as pose estimation, visual localization, and sparse 3D reconstruction.

Visual simultaneous localisation and mapping (vSLAM) finds applications for indoor and outdoor navigation that routinely subjects it to visual complexities, particularly mirror reflections. The effect of mirror presence (time visible and its average size in the frame) was hypothesised to impact localisation and mapping performance, with systems using direct techniques expected to perform worse. Thus, a dataset, MirrEnv, of image sequences recorded in mirror environments, was collected, and used to evaluate the performance of existing representative methods. RGBD ORB-SLAM3 and BundleFusion appear to show moderate degradation of absolute trajectory error with increasing mirror duration, whilst the remaining results did not show significantly degraded localisation performance. The mesh maps generated proved to be very inaccurate, with real and virtual reflections colliding in the reconstructions. A discussion is given of the likely sources of error and robustness in mirror environments, outlining future directions for validating and improving vSLAM performance in the presence of planar mirrors. The MirrEnv dataset is available at https://doi.org/10.17035/d.2023.0292477898 .

Irregular boundaries in image stitching naturally occur due to freely moving cameras. To deal with this problem, existing methods focus on optimizing mesh warping to make boundaries regular using the traditional explicit solution. However, previous methods always depend on hand-crafted features (e.g., keypoints and line segments). Thus, failures often happen in overlapping regions without distinctive features. In this paper, we address this problem by proposing RecStitchNet, a reasonable and effective network for image stitching with rectangular boundaries. Considering that both stitching and imposing rectangularity are non-trivial tasks in the learning-based framework, we propose a three-step progressive learning based strategy, which not only simplifies this task, but gradually achieves a good balance between stitching and imposing rectangularity. In the first step, we perform initial stitching by a pre-trained state-of-the-art image stitching model, to produce initially warped stitching results without considering the boundary constraint. Then, we use a regression network with a comprehensive objective regarding mesh, perception, and shape to further encourage the stitched meshes to have rectangular boundaries with high content fidelity. Finally, we propose an unsupervised instance-wise optimization strategy to refine the stitched meshes iteratively, which can effectively improve the stitching results in terms of feature alignment, as well as boundary and structure preservation. Due to the lack of stitching datasets and the difficulty of label generation, we propose to generate a stitching dataset with rectangular stitched images as pseudo-ground-truth labels, and the performance upper bound induced from the it can be broken by our unsupervised refinement. Qualitative and quantitative results and evaluations demonstrate the advantages of our method over the state-of-the-art.

Picking up an entangled object is a difficult manipulation task due to its rich contact dynamics. Most existing solutions fail to produce grasp poses to enable reliable manipulation due to the dependence on simplified assumptions for the motion policies. Grasps generated by these methods tend to drop objects or cause undesired movements of non-grasped objects. To improve such object-disentangling tasks, we propose to extend the concept of reinforcement learning (RL)-based affordance to include arbitrary action consequences and implement a general affordance-based manipulation (GAM) framework. In the GAM, we train an RL agent that uses more fine-grained actions and outperforms previous methods with a smaller chance of dropping objects and making contact with non-grasped hooks. Then, a manipulation affordance prediction (MAP) model is trained to estimate the performances of the RL agent. Finally, the manipulation affordance-based grasp filter (MAGF) selects grasp poses that afford the desired manipulation performances, showing substantial improvements in five challenging hook disentangling tasks in simulation. The experiments show (1) the limitation of TAG generators, (2) the effectiveness of filtering TAGs with predicted manipulation performances based on the general affordance theory, and (3) the importance of avoiding contact with non-grasped objects in contact-rich manipulation.

Texture synthesis is a fundamental problem in computer graphics that would benefit various applications. Existing methods are effective in handling 2D image textures. In contrast, many real-world textures contain meso-structure in the 3D geometry space, such as grass, leaves, and fabrics, which cannot be effectively modeled using only 2D image textures. We propose a novel texture synthesis method with Neural Radiance Fields (NeRF) to capture and synthesize textures from given multi-view images. In the proposed NeRF texture representation, a scene with fine geometric details is disentangled into the meso-structure textures and the underlying base shape. This allows textures with meso-structure to be effectively learned as latent features situated on the base shape, which are fed into a NeRF decoder trained simultaneously to represent the rich view-dependent appearance. Using this implicit representation, we can synthesize NeRF-based textures through patch matching of latent features. However, inconsistencies between the metrics of the reconstructed content space and the latent feature space may compromise the synthesis quality. To enhance matching performance, we further regularize the distribution of latent features by incorporating a clustering constraint. In addition to generating NeRF textures over a planar domain, our method can also synthesize NeRF textures over curved surfaces, which are practically useful. Experimental results and evaluations demonstrate the effectiveness of our approach.

Emotions can be perceived from a person's gait, i.e., their walking style. Existing methods on gait emotion recognition mainly leverage the posture information as input, but ignore the body movement, which contains complementary information for recognizing emotions evoked in the gait. In this paper, we propose a Bilateral Posture and Movement Graph Convolutional Network (BPM-GCN) that consists of two parallel streams, namely posture stream and movement stream, to recognize emotions from two views. The posture stream aims to explicitly analyse the emotional state of the person. Specifically, we design a novel regression constraint based on the hand-engineered features to distill the prior affective knowledge into the network and boost the representation learning. The movement stream is designed to describe the intensity of the emotion, which is an implicitly cue for recognizing emotions. To achieve this goal, we employ a higher-order velocity-acceleration pair to construct graphs, in which the informative movement features are utilized. Besides, we design a PM-Interacted feature fusion mechanism to adaptively integrate the features from the two streams. Therefore, the two streams collaboratively contribute to the performance from two complementary views. Extensive experiments on the largest benchmark dataset Emotion-Gait show that BPM-GCN performs favorably against the state-of-the-art approaches (with at least$4.59\%$performance improvement). The source code is released onhttps://github.com/exped1230/BPM-GCN.

Much progress has been made in reconstructing garments from an image or a video. However, none of existing works meet the expectations of digitizing high-quality animatable dynamic garments that can be adjusted to various unseen poses. In this paper, we propose the first method to recover high-quality animatable dynamic garments from monocular videos without depending on scanned data. To generate reasonable deformations for various unseen poses, we propose a learnable garment deformation network that formulates the garment reconstruction task as a pose-driven deformation problem. To alleviate the ambiguity estimating 3D garments from monocular videos, we design a multi-hypothesis deformation module that learns spatial representations of multiple plausible deformations. Experimental results on several public datasets demonstrate that our method can reconstruct high-quality dynamic garments with coherent surface details, which can be easily animated under unseen poses. The code will be provided for research purposes.

Group detection, especially for large-scale scenes, has many potential applications for public safety and smart cities. Existing methods fail to cope with frequent occlusions in large-scale scenes with multiple people, and are difficult to effectively utilize spatio-temporal information. In this paper, we propose an end-to-end framework,GroupTransformer, for group detection in large-scale scenes. To deal with the frequent occlusions caused by multiple people, we design an occlusion encoder to detect and suppress severely occluded person crops. To explore the potential spatio-temporal relationship, we propose spatio-temporal transformers to simultaneously extract trajectory information and fuse inter-person features in a hierarchical manner. Experimental results on both large-scale and small-scale scenes demonstrate that our method achieves better performance compared with state-of-the-art methods. On large-scale scenes, our method significantly boosts the performance in terms of precision and F1 score by more than 10%. On small-scale scenes, our method still improves the performance of F1 score by more than 5%.We will release the code for research purposes.

2D-3D joint learning is essential and effective for fundamental 3D vision tasks, such as 3D semantic segmentation, due to the complementary information these two visual modalities contain. Most current 3D scene semantic segmentation methods process 2D images "as they are", i.e., only real captured 2D images are used. However, such captured 2D images may be redundant, with abundant occlusion and/or limited field of view (FoV), leading to poor performance for the current methods involving 2D inputs. In this paper, we propose a general learning framework for joint 2D-3D scene understanding by selecting informative virtual 2D views of the underlying 3D scene. We then feed both the 3D geometry and the generated virtual 2D views into any joint 2D-3D-input or pure 3D-input based deep neural models for improving 3D scene understanding. Specifically, we generate virtual 2D views based on an information score map learned from the current 3D scene semantic segmentation results. To achieve this, we formalize the learning of the information score map as a deep reinforcement learning process, which rewards good predictions using a deep neural network. To obtain a compact set of virtual 2D views that jointly cover informative surfaces of the 3D scene as much as possible, we further propose an efficient greedy virtual view coverage strategy in the normal-sensitive 6D space, including 3-dimensional point coordinates and 3-dimensional normal. We have validated our proposed framework for various joint 2D-3D-input or pure 3D-input based deep neural models on two real-world 3D scene datasets, i.e., ScanNet v2 and S3DIS, and the results demonstrate that our method obtains a consistent gain over baseline models and achieves new top accuracy for joint 2D and 3D scene semantic segmentation. Code is available at https://github.com/smy-THU/VirtualViewSelection.

Motion models are vital for solving multiple object tracking (MOT), which makes instance-level position predictions of targets to handle occlusions and noisy detections. Recent methods have proposed the use of Single Object Tracking (SOT) techniques to build motion models and unify the SOT tracker with the object detector into a single network for high-efficiency MOT. However, three feature incompatibility issues in the required features of this paradigm are ignored, leading to inferior performance. First, the object detector requires class-specific features to localize objects of pre-defined classes. Contrarily, target-specific features are required in SOT to track the target of interest with an unknown category. Second, MOT relies on intra-class differences to associate targets of the same identity (ID). On the other hand, the SOT trackers focus on inter-class differences to distinguish the tracking target from the background. Third, classification confidence is used to determine the existence of targets, which is obtained with category-related features and cannot accurately reveal the existence of targets in tracking scenes. To address these issues, we propose a novel Task-specific Feature Encoding Network (TFEN) to extract task-driven features for different sub-networks. Besides, we propose a novel Quadruplet State Sampling (QSS) strategy to form the training samples of the motion model and guide the SOT trackers to capture identity-discriminative features in position predictions. Finally, we propose an Existence Aware Tracking (EAT) algorithm by estimating the existence confidence of targets and re-considering low-scored predictions to recover missed targets. Experimental results indicate that the proposed Discriminative Motion Model-based tracker (DMMTracker) can effectively address these issues when employing SOT trackers as motion models, leading to highly competitive results on MOT benchmarks.

Text-driven image editing aims to manipulate images with the guidance of natural language description. Text is much more natural and intuitive than many other interaction modes, and attracts more attention recently. However, compared with classical supervised learning tasks, there is no standard benchmark dataset for text-driven interactive image editing up to now. Therefore, it is hard to train an end-to-end model for pixel-aligned interactive image editing driven by text. Some methods follow the paradigm of text-to-image models by incorporating the target image into the process of text-to-image generation. However, these methods relying on cross-modal text-to-image generation involve complicated and expensive models, which can lead to inconsistent editing effects. In this article, a novel text-driven image editing method is proposed. Our key observation is that this task can be more efficiently learned using image-to-image translation. To ensure effective learning for image editing, our framework takes paired text and the corresponding images for training, and disentangles each image into content and attributes, such that the content is maintained while the attributes are modified according to the text. Our network is a lightweight encoder-decoder architecture that accomplishes pixel-aligned end-to-end training via cycle-consistent supervision. Quantitative and qualitative experimental results show that the proposed method achieves state-of-the-art performance.

Neural implicit representations, including Neural Distance Fields and Neural Radiance Fields, have demonstrated significant capabilities for reconstructing surfaces with complicated geometry and topology, and generating novel views of a scene. Nevertheless, it is challenging for users to directly deform or manipulate these implicit representations with large deformations in a real-time fashion. Gaussian Splatting (GS) has recently become a promising method with explicit geometry for representing static scenes and facilitating high-quality and real-time synthesis of novel views. However, it cannot be easily deformed due to the use of discrete Gaussians and the lack of explicit topology. To address this, we develop a novel GS-based method (GaussianMesh) that enables interactive deformation. Our key idea is to design an innovative mesh-based GS representation, which is integrated into Gaussian learning and manipulation. 3D Gaussians are defined over an explicit mesh, and they are bound with each other: the rendering of 3D Gaussians guides the mesh face split for adaptive refinement, and the mesh face split directs the splitting of 3D Gaussians. Moreover, the explicit mesh constraints help regularize the Gaussian distribution, suppressing poor-quality Gaussians ( e.g. , misaligned Gaussians, long-narrow shaped Gaussians), thus enhancing visual quality and reducing artifacts during deformation. Based on this representation, we further introduce a large-scale Gaussian deformation technique to enable deformable GS, which alters the parameters of 3D Gaussians according to the manipulation of the associated mesh. Our method benefits from existing mesh deformation datasets for more realistic data-driven Gaussian deformation. Extensive experiments show that our approach achieves high-quality reconstruction and effective deformation, while maintaining the promising rendering results at a high frame rate (65 FPS on average on a single commodity GPU).

Existing text-based 3D generation methods generate attractive results but lack detailed geometry control. Sketches, known for their conciseness and expressiveness, have contributed to intuitive 3D modeling but are confined to producing texture-less mesh models within predefined categories. Integrating sketch and text simultaneously for 3D generation promises enhanced control over geometry and appearance but faces challenges from 2D-to-3D translation ambiguity and multi-modal condition integration. Moreover, further editing of 3D models in arbitrary views will give users more freedom to customize their models. However, it is difficult to achieve high generation quality, preserve unedited regions, and manage proper interactions between shape components. To solve the above issues, we propose a text-driven 3D content generation and editing method, SketchDream, which supports NeRF generation from given hand-drawn sketches and achieves free-view sketch-based local editing. To tackle the 2D-to-3D ambiguity challenge, we introduce a sketch-based multi-view image generation diffusion model, which leverages depth guidance to establish spatial correspondence. A 3D ControlNet with a 3D attention module is utilized to control multi-view images and ensure their 3D consistency. To support local editing, we further propose a coarse-to-fine editing approach: the coarse phase analyzes component interactions and provides 3D masks to label edited regions, while the fine stage generates realistic results with refined details by local enhancement. Extensive experiments validate that our method generates higher-quality results compared with a combination of 2D ControlNet and image-to-3D generation techniques and achieves detailed control compared with existing diffusion-based 3D editing approaches.

3D scene synthesis facilitates and benefits many real-world applications. Most scene generators focus on making indoor scenes plausible via learning from training data and leveraging extra constraints such as adjacency and symmetry. Although the generated 3D scenes are mostly plausible with visually realistic layouts, they can be functionally unsuitable for human users to navigate and interact with furniture. Our key observation is that human activity plays a critical role and sufficient free space is essential for human-scene interactions. This is exactly where many existing synthesized scenes fail—the seemingly correct layouts are often not fit for living. To tackle this, we present a human-aware optimization framework Haisor for 3D indoor scene arrangement via reinforcement learning, which aims to find an action sequence to optimize the indoor scene layout automatically. Based on the hierarchical scene graph representation, an optimal action sequence is predicted and performed via Deep Q-Learning with Monte Carlo Tree Search (MCTS), where MCTS is our key feature to search for the optimal solution in long-term sequences and large action space. Multiple human-aware rewards are designed as our core criteria of human-scene interaction, aiming to identify the next smart action by leveraging powerful reinforcement learning. Our framework is optimized end-to-end by giving the indoor scenes with part-level furniture layout including part mobility information. Furthermore, our methodology is extensible and allows utilizing different reward designs to achieve personalized indoor scene synthesis. Extensive experiments demonstrate that our approach optimizes the layout of 3D indoor scenes in a human-aware manner, which is more realistic and plausible than original state-of-the-art generator results, and our approach produces superior smart actions, outperforming alternative baselines.

As communications are increasingly taking place virtually, the ability to present well online is becoming an indispensable skill. Online speakers are facing unique challenges in engaging with remote audiences. However, there has been a lack of evidence-based analytical systems for people to comprehensively evaluate online speeches and further discover possibilities for improvement. This paper introduces SpeechMirror, a visual analytics system facilitating reflection on a speech based on insights from a collection of online speeches. The system estimates the impact of different speech techniques on effectiveness and applies them to a speech to give users awareness of the performance of speech techniques. A similarity recommendation approach based on speech factors or script content supports guided exploration to expand knowledge of presentation evidence and accelerate the discovery of speech delivery possibilities. SpeechMirror provides intuitive visualizations and interactions for users to understand speech factors. Among them, SpeechTwin, a novel multimodal visual summary of speech, supports rapid understanding of critical speech factors and comparison of different speech samples, and SpeechPlayer augments the speech video by integrating visualization of the speaker's body language with interaction, for focused analysis. The system utilizes visualizations suited to the distinct nature of different speech factors for user comprehension. The proposed system and visualization techniques were evaluated with domain experts and amateurs, demonstrating usability for users with low visualization literacy and its efficacy in assisting users to develop insights for potential improvement.

Background \nWith the development of virtual reality (VR) technology, there is a growing need for customized 3D avatars. However, traditional methods for 3D avatar modeling are either time-consuming or fail to retain the similarity to the person being modeled. This study presents a novel framework for generating animatable 3D cartoon faces from a single portrait image. \n \nMethods \nFirst, we transferred an input real-world portrait to a stylized cartoon image using StyleGAN. We then proposed a two-stage reconstruction method to recover a 3D cartoon face with detailed texture. Our two-stage strategy initially performs coarse estimation based on template models and subsequently refines the model by nonrigid deformation under landmark supervision. Finally, we proposed a semantic-preserving face-rigging method based on manually created templates and deformation transfer. \n \nConclusions \nCompared with prior arts, the qualitative and quantitative results show that our method achieves better accuracy, aesthetics, and similarity criteria. Furthermore, we demonstrated the capability of the proposed 3D model for real-time facial animation.

Generative Adversarial networks (GANs) have demonstrated their powerful capability of synthesizing high-resolution images, and great efforts have been made to interpret the semantics in the latent spaces of GANs. However, existing works still have the following limitations: (1) the majority of works rely on either pretrained attribute predictors or large-scale labeled datasets, which are difficult to collect in most cases, and (2) some other methods are only suitable for restricted cases, such as focusing on interpretation of human facial images using prior facial semantics. In this paper, we propose a GAN-based method called FEditNet, aiming to discover latent semantics using very few labeled data without any pretrained predictors or prior knowledge. Specifically, we reuse the knowledge from the pretrained GANs, and by doing so, avoid overfitting during the few-shot training of FEditNet. Moreover, our layer-wise objectives which take content consistency into account also ensure the disentanglement between attributes. Qualitative and quantitative results demonstrate that our method outperforms the state-of-the-art methods on various datasets. The code is available at https://github.com/THU-LYJ-Lab/FEditNet.

Image-to-image translation is an important and challenging problem in computer vision and image processing. Diffusion models (DM) have shown great potentials for high-quality image synthesis, and have gained competitive performance on the task of image-to-image translation. However, most of the existing diffusion models treat image-to-image translation as conditional generation processes, and suffer heavily from the gap between distinct domains. In this paper, a novel image-to-image translation method based on the Brownian Bridge Diffusion Model (BBDM) is proposed, which models image-to-image translation as a stochastic Brownian Bridge process, and learns the translation between two domains directly through the bidirectional diffusion process rather than a conditional generation process. To the best of our knowledge, it is the first work that proposes Brownian Bridge diffusion process for image-to-image translation. Experimental results on various benchmarks demonstrate that the proposed BBDM model achieves competitive performance through both visual inspection and measurable metrics.

3D human body representation learning has received increasing attention in recent years. However, existing works cannot flexibly, controllably and accurately represent human bodies, limited by coarse semantics and unsatisfactory representation capability, particularly in the absence of supervised data. In this paper, we propose a human body representation with fine-grained semantics and high reconstruction-accuracy in an unsupervised setting. Specifically, we establish a correspondence between latent vectors and geometric measures of body parts by designing a part-aware skeleton-separated decoupling strategy, which facilitates controllable editing of human bodies by modifying the corresponding latent codes. With the help of a bone-guided auto-encoder and an orientation-adaptive weighting strategy, our representation can be trained in an unsupervised manner. With the geometrically meaningful latent space, it can be applied to a wide range of applications, from human body editing to latent code interpolation and shape style transfer. Experimental results on public datasets demonstrate the accurate reconstruction and flexible editing abilities of the proposed method. The code will be available at http://cic.tju.edu.cn/faculty/likun/projects/SemanticHuman.

Image-based multi-person reconstruction in wide-field large scenes is critical for crowd analysis and security alert. However, existing methods cannot deal with large scenes containing hundreds of people, which encounter the challenges of large number of people, large variations in human scale, and complex spatial distribution. In this paper, we propose Crowd3D, the first framework to reconstruct the 3D poses, shapes and locations of hundreds of people with global consistency from a single large-scene image. The core of our approach is to convert the problem of complex crowd localization into pixel localization with the help of our newly defined concept, Human-scene Virtual Interaction Point (HVIP). To reconstruct the crowd with global consistency, we propose a progressive reconstruction network based on HVIP by pre-estimating a scene-level camera and a ground plane. To deal with a large number of persons and various human sizes, we also design an adaptive human-centric cropping scheme. Besides, we contribute a benchmark dataset, LargeCrowd, for crowd reconstruction in a large scene. Experimental results demonstrate the effectiveness of the proposed method. The code and the dataset are available at http://cic.tju.edu.cn/faculty/likun/projects/Crowd3D.

3D single object tracking plays an essential role in many applications, such as autonomous driving. It remains a challenging problem due to the large appearance variation and the sparsity of points caused by occlusion and lim-ited sensor capabilities. Therefore, contextual information across two consecutive frames is crucial for effective object tracking. However, points containing such useful information are often overlooked and cropped out in existing methods, leading to insufficient use of important contextual knowledge. To address this issue, we propose CXTrack, a novel transformer-based network for 3D object tracking, which exploits ConteXtual information to improve the tracking results. Specifically, we design a target-centric transformer network that directly takes point features from two consecutive frames and the previous bounding box as input to explore contextual information and implicitly propagate target cues. To achieve accurate localization for objects of all sizes, we propose a transformer-based localization head with a novel center embedding module to distinguish the target from distractors. Extensive experiments on three large-scale datasets, KITTI, nuScenes and Waymo Open Dataset, show that CXTrack achieves state-of-the-art tracking performance while running at 34 FPS.

Image aesthetics assessment (IAA) is a challenging task due to its highly subjective nature. Most of the current studies rely on large-scale datasets (e.g., AVA and AADB) to learn a general model for all kinds of photography images. However, little light has been shed on measuring the aesthetic quality of artistic images, and the existing datasets only contain relatively few artworks. Such a defect is a great obstacle to the aesthetic assessment of artistic images. To fill the gap in the field of artistic image aesthetics assessment (AIAA), we first introduce a large-scale AIAA dataset: Boldbrush Artistic Image Dataset (BAlD), which consists of 60,337 artistic images covering various art forms, with more than 360,000 votes from online users. We then propose a new method, SAAN (Style-specific Art Assessment Network), which can effectively extract and utilize style-specific and generic aesthetic information to evaluate artistic images. Experiments demonstrate that our proposed approach outperforms existing lAA methods on the proposed BAlD dataset according to quantitative comparisons. We believe the proposed dataset and method can serve as a foundation for future AIAA works and inspire more research in this field. Dataset and code are available at: https://github.com/Dreemurr-T/BAID.git

Detecting symmetrical properties is a fundamental task in 3D shape analysis. In the case of a 3D model with planar symmetries, each point has a corresponding mirror point w.r.t. a symmetry plane, and the correspondences remain invariant under any arbitrary Euclidean transformation. Our proposed method, E3Sym, aims to detect planar reflective symmetry in an unsupervised and end-to-end manner by leveraging E(3) invariance. E3Sym establishes robust point correspondences through the use of E(3) invariant features extracted from a lightweight neural network, from which the dense symmetry prediction is produced. We also introduce a novel and efficient clustering algorithm to aggregate the dense prediction and produce a detected symmetry set, allowing for the detection of an arbitrary number of planar symmetries while ensuring the method remains differentiable for end-to-end training. Our method also possesses the ability to infer reasonable planar symmetries from incomplete shapes, which remains challenging for existing methods. Extensive experiments demonstrate that E3Sym is both effective and robust, outperforming state-of-the-art methods.

Despite the success of StyleGAN in image synthesis, the images it synthesizes are not always perfect and the well-known truncation trick has become a standard post-processing technique for StyleGAN to synthesize high quality images. Although effective, it has long been noted that the truncation trick tends to reduce the diversity of synthesized images and unnecessarily sacrifices many distinct image features. To address this issue, in this paper, we first delve into the StyleGAN image synthesis mechanism and discover an important phenomenon, namely Feature Proliferation, which demonstrates how specific features reproduce with forward propagation. Then, we show how the occurrence of Feature Proliferation results in StyleGAN image artifacts. As an analogy, we refer to it as the "cancer" in StyleGAN from its proliferating and malignant nature. Finally, we propose a novel feature rescaling method that identifies and modulates risky features to mitigate feature proliferation. Thanks to our discovery of Feature Proliferation, the proposed feature rescaling method is less destructive and retains more useful image features than the truncation trick, as it is more fine-grained and works in a lower-level feature space rather than a high-level latent space. Experimental results justify the validity of our claims and the effectiveness of the proposed feature rescaling method. Our code is available at https://github.com/songc42/Feature-proliferation.

3D single object tracking has been a crucial problem for decades with numerous applications such as autonomous driving. Despite its wide-ranging use, this task remains challenging due to the significant appearance variation caused by occlusion and size differences among tracked targets. To address these issues, we present MBPTrack, which adopts a Memory mechanism to utilize past information and formulates localization in a coarse-to-fine scheme using Box Priors given in the first frame. Specifically, past frames with targetness masks serve as an extenral memory, and a transformer-based module propagates tracked target cues from the memory to the current frame. To precisely localize objects of all sizes, MBPTrack first predicts the target center via Hough voting. By leveraging box priors given in the first frame, we adaptively sample reference points around the target center that roughly cover the target of different sizes. Then, we obtain dense feature maps by aggregating point features into the reference points, where localization can be performed more effectively. Extensive experiments demonstrate that MBPTrack achieves state-of-the-art performance on KITTI, nuScenes and Waymo Open Dataset, while running at 50 FPS on a single RTX3090 GPU.

Learning-based high-fidelity reconstruction of 3D shapes with varying topology is a fundamental problem in computer vision and computer graphics. Recent advances in learning 3D shapes using explicit and implicit representations have achieved impressive results in 3D modeling. However, the template-based explicit representation is limited by fixed topology, and the implicit representation, although flexible with arbitrary topology, requires a large number of sampled points to regress the surface, which is computationally expensive. In this work, we propose a novel 3D shape representation named NeuralSlice, which represents a 3D shape as the intersection of a 4D tetrahedral mesh and a 4D hyperplane. A novel network is designed to incorporate the proposed representation flexibly, which learns a deformable 4D template and a parameter for slicing 4D hyperplane to reconstruct the 3D object. To learn the local deformation of the 4D template, we further propose a spatial-aware network to locate the 4D points within the 3D feature volume of input shape via positional encoding, which leverages the local geometrical feature to guide the 4D deformation. By addressing the 3D problem in a higher 4D space, our method supports flexible topology changes while being highly efficient. Our method is guaranteed to produce manifold meshes. NeuralSlice outperforms the state-of-the-art explicit-based approaches in terms of reconstruction quality. Compared with implicit approaches, by avoiding point sampling, our method is 10 times faster than the implicit approaches, and better preserves thin structures. NeuralSlice has the capability of representing various shapes and topologies using a single 4D tetrahedral mesh. The corresponding code can be found on GitHub at https://github.com/IGLICT/NEURALSLICE

Outdoor scenes often involve large-scale geometry and complex unknown lighting conditions, making it difficult to decompose them into geometry, reflectance and illumination. Recently researchers made attempts to decompose outdoor scenes using Neural Radiance Fields (NeRF) and learning-based lighting and shadow representations. However, diverse lighting conditions and shadows in outdoor scenes are challenging for learning-based models. Moreover, existing methods may produce rough geometry and normal reconstruction and introduce notable shading artifacts when the scene is rendered under a novel illumination. To solve the above problems, we propose SOL-NeRF to decompose outdoor scenes with the help of a hybrid lighting representation and a signed distance field geometry reconstruction. We use a single Spherical Gaussian (SG) lobe to approximate the sun lighting, and a first-order Spherical Harmonic (SH) mixture to resemble the sky lighting. This hybrid representation is specifically designed for outdoor settings, and compactly models the outdoor lighting, ensuring robustness and efficiency. The shadow of the direct sun lighting can be obtained by casting the ray against the mesh extracted from the signed distance field, and the remaining shadow can be approximated by Ambient Occlusion (AO). Additionally, sun lighting color prior and a relaxed Manhattan-world assumption can be further applied to boost decomposition and relighting performance. When changing the lighting condition, our method can produce consistent relighting results with correct shadow effects. Experiments conducted on our hybrid lighting scheme and the entire decomposition pipeline show that our method achieves better reconstruction, decomposition, and relighting performance compared to previous methods both quantitatively and qualitatively.

In this paper, we propose a novel framework, Tracking-free Relightable Avatar (TRAvatar), for capturing and reconstructing high-fidelity 3D avatars. Compared to previous methods, TRAvatar works in a more practical and efficient setting. Specifically, TRAvatar is trained with dynamic image sequences captured in a Light Stage under varying lighting conditions, enabling realistic relighting and real-time animation for avatars in diverse scenes. Additionally, TRAvatar allows for tracking-free avatar capture and obviates the need for accurate surface tracking under varying illumination conditions. Our contributions are two-fold: First, we propose a novel network architecture that explicitly builds on and ensures the satisfaction of the linear nature of lighting. Trained on simple group light captures, TRAvatar can predict the appearance in real-time with a single forward pass, achieving high-quality relighting effects under illuminations of arbitrary environment maps. Second, we jointly optimize the facial geometry and relightable appearance from scratch based on image sequences, where the tracking is implicitly learned. This tracking-free approach brings robustness for establishing temporal correspondences between frames under different lighting conditions. Extensive qualitative and quantitative experiments demonstrate that our framework achieves superior performance for photorealistic avatar animation and relighting.

Multi-class part parsing is a dense prediction task that seeks to simultaneously detect multiple objects and the semantic parts within these objects in the scene. This problem is important in providing detailed object understanding, but is challenging due to the existence of both class-level and part-level ambiguities. In this paper, we propose to integrate an attention refinement module and a feature fusion module to tackle the part-level ambiguity. The attention refinement module aims to enhance the feature representations by focusing on important features. The feature fusion module aims to improve the fusion operation for different scales of features. We also propose an object-to-part training strategy to tackle the class-level ambiguity, which improves the localization of parts by exploiting prior knowledge of objects. The experimental results demonstrated the effectiveness of the proposed modules and the training strategy, and showed that our proposed method achieved state-of-the-art performance on the benchmark datasets.

Depth images can be easily acquired using depth cameras. However, these images only contain partial information about the shape due to unavoidable self-occlusion. Thanks to the availability of large datasets of shapes, it is possible to use a learning-based approach to produce complete shapes from single depth images. State-of-the-art generative adversarial network (GAN) architectures can produce reasonable results. However, the use of relatively local convolutions restricts GAN architectures from producing globally plausible shapes. In this study, we develop a novel dynamic latent code selection mechanism in which the model learns to select only important codes from the latent space. Furthermore, a novel 3D self-attention (3DSA) layer is introduced that is able to capture non-local relationships across the 3D space. We further design a GAN architecture that uses a multistage encoder–decoder to recover the shape, where our 3DSA layer is introduced to the discriminator to help attend to global features, which stabilizes the model learning and encourages shape refinement, making our reconstruction more structurally plausible. Through extensive experiments, we demonstrate that our method outperforms other state-of-the-art methods for single depth image 3D reconstruction.

Abstract In physics‐based cloth animation, rich folds and detailed wrinkles are achieved at the cost of expensive computational resources and huge labor tuning. Data‐driven techniques make efforts to reduce the computation significantly by utilizing a preprocessed database. One type of methods relies on human poses to synthesize fitted garments, but these methods cannot be applied to general cloth animations. Another type of methods adds details to the coarse meshes obtained through simulation, which does not have such restrictions. However, existing works usually utilize coordinate‐based representations which cannot cope with large‐scale deformation, and requires dense vertex correspondences between coarse and fine meshes. Moreover, as such methods only add details, they require coarse meshes to be sufficiently close to fine meshes, which can be either impossible, or require unrealistic constraints to be applied when generating fine meshes. To address these challenges, we develop a temporally and spatially as‐consistent‐as‐possible deformation representation (named TS‐ACAP) and design a DeformTransformer network to learn the mapping from low‐resolution meshes to ones with fine details. This TS‐ACAP representation is designed to ensure both spatial and temporal consistency for sequential large‐scale deformations from cloth animations. With this TS‐ACAP representation, our DeformTransformer network first utilizes two mesh‐based encoders to extract the coarse and fine features using shared convolutional kernels, respectively. To transduct the coarse features to the fine ones, we leverage the spatial and temporal Transformer network that consists of vertex‐level and frame‐level attention mechanisms to ensure detail enhancement and temporal coherence of the prediction. Experimental results show that our method is able to produce reliable and realistic animations in various datasets at high frame rates with superior detail synthesis abilities compared to existing methods.

Novel viewpoint image synthesis is very challenging, especially from sparse views, due to large changes in viewpoint and occlusion. Existing image-based methods fail to generate reasonable results for invisible regions, while geometry-based methods have difficulties in synthesizing detailed textures. In this paper, we propose STATE, an end-to-end deep neural network, for sparse view synthesis by learning structure and texture representations. Structure is encoded as a hybrid feature field to predict reasonable structures for invisible regions while maintaining original structures for visible regions, and texture is encoded as a deformed feature map to preserve detailed textures. We propose a hierarchical fusion scheme with intra-branch and inter-branch aggregation, in which spatio-view attention allows multi-view fusion at the feature level to adaptively select important information by regressing pixel-wise or voxel-wise confidence maps. By decoding the aggregated features, STATE is able to generate realistic images with reasonable structures and detailed textures. Experimental results demonstrate that our method achieves qualitatively and quantitatively better results than state-of-the-art methods. Our method also enables texture and structure editing applications benefiting from implicit disentanglement of structure and texture. Our code is available at http://cic.tju.edu.cn/faculty/likun/projects/STATE .

3D indoor scenes are widely used in computer graphics, with applications ranging from interior design to gaming to virtual and augmented reality. They also contain rich information, including room layout, as well as furniture type, geometry, and placement. High-quality 3D indoor scenes are highly demanded while it requires expertise and is time-consuming to design high-quality 3D indoor scenes manually. Existing research only addresses partial problems: some works learn to generate room layout, and other works focus on generating detailed structure and geometry of individual furniture objects. However, these partial steps are related and should be addressed together for optimal synthesis. We propose SceneHGN, a hierarchical graph network for 3D indoor scenes that takes into account the full hierarchy from the room level to the object level, then finally to the object part level. Therefore for the first time, our method is able to directly generate plausible 3D room content, including furniture objects with fine-grained geometry, and their layout. To address the challenge, we introduce functional regions as intermediate proxies between the room and object levels to make learning more manageable. To ensure plausibility, our graph-based representation incorporates both vertical edges connecting child nodes with parent nodes from different levels, and horizontal edges encoding relationships between nodes at the same level. Our generation network is a conditional recursive neural network (RvNN) based variational autoencoder (VAE) that learns to generate detailed content with fine-grained geometry for a room, given the room boundary as the condition. Extensive experiments demonstrate that our method produces superior generation results, even when comparing results of partial steps with alternative methods that can only achieve these. We also demonstrate that our method is effective for various applications such as part-level room editing, room interpolation, and room generation by arbitrary room boundaries.

We propose a new method for realistic human motion transfer using a generative adversarial network (GAN), which generates a motion video of a target character imitating actions of a source character, while maintaining high authenticity of the generated results. We tackle the problem by decoupling and recombining the posture information and appearance information of both the source and target characters. The innovation of our approach lies in the use of the projection of a reconstructed 3D human model as the condition of GAN to better maintain the structural integrity of transfer results in different poses. We further introduce a detail enhancement net to enhance the details of transfer results by exploiting the details in real source frames. Extensive experiments show that our approach yields better results both qualitatively and quantitatively than the state-of-the-art methods.

Pose transfer of human videos aims to generate a high-fidelity video of a target person imitating actions of a source person. A few studies have made great progress either through image translation with deep latent features or neural rendering with explicit 3D features. However, both of them rely on large amounts of training data to generate realistic results, and the performance degrades on more accessible Internet videos due to insufficient training frames. In this paper, we demonstrate that the dynamic details can be preserved even when trained from short monocular videos. Overall, we propose a neural video rendering framework coupled with an image-translation-based dynamic details generation network (D$^{2}$G-Net), which fully utilizes both the stability of explicit 3D features and the capacity of learning components. To be specific, a novel hybrid texture representation is presented to encode both the static and pose-varying appearance characteristics, which is then mapped to the image space and rendered as a detail-rich frame in the neural rendering stage. Through extensive comparisons, we demonstrate that our neural human video renderer is capable of achieving both clearer dynamic details and more robust performance even on accessible short videos with only 2 k$\sim$4 k frames, as illustrated in Fig. 1.

We present a novel method for single-view 3D reconstruction of textured meshes, with a focus to address the primary challenge surrounding texture inference and transfer. Our key observation is that learning textured reconstruction in a structure-aware and globally consistent manner is effective in handling the severe ill-posedness of the texturing problem and significant variations in object pose and texture details. Specifically, we perform structured mesh reconstruction, via a retrieval-and-assembly approach, to produce a set of genus-zero parts parameterized by deformable boxes and endowed with semantic information. For texturing, we first transfer visible colors from the input image onto the unified UV texture space of the deformable boxes. Then we combine a learned transformer model for per-part texture completion with a global consistency loss to optimize inter-part texture consistency. Our texture completion model operates in a VQ-VAE embedding space and is trained end-to-end, with the transformer training enhanced with retrieved texture instances to improve texture completion performance amid significant occlusion. Extensive experiments demonstrate higher-quality textured mesh reconstruction obtained by our method over state-of-the-art alternatives, both quantitatively and qualitatively, as reflected by a better recovery of texture coherence and details.

Face portrait line drawing is a unique style of art which is highly abstract and expressive. However, due to its high semantic constraints, many existing methods learn to generate portrait drawings using paired training data, which is costly and time-consuming to obtain. In this paper, we propose a novel method to automatically transform face photos to portrait drawings using unpaired training data with two new features; i.e., our method can (1) learn to generate high quality portrait drawings in multiple styles using a single network and (2) generate portrait drawings in a "new style" unseen in the training data. To achieve these benefits, we (1) propose a novel quality metric for portrait drawings which is learned from human perception, and (2) introduce a quality loss to guide the network toward generating better looking portrait drawings. We observe that existing unpaired translation methods such as CycleGAN tend to embed invisible reconstruction information indiscriminately in the whole drawings due to significant information imbalance between the photo and portrait drawing domains, which leads to important facial features missing. To address this problem, we propose a novel asymmetric cycle mapping that enforces the reconstruction information to be visible and only embedded in the selected facial regions. Along with localized discriminators for important facial regions, our method well preserves all important facial features in the generated drawings. Generator dissection further explains that our model learns to incorporate face semantic information during drawing generation. Extensive experiments including a user study show that our model outperforms state-of-the-art methods.

The recently proposed neural radiance fields (NeRF) use a continuous function formulated as a multi-layer perceptron (MLP) to model the appearance and geometry of a 3D scene. This enables realistic synthesis of novel views, even for scenes with view dependent appearance. Many follow-up works have since extended NeRFs in different ways. However, a fundamental restriction of the method remains that it requires a large number of images captured from densely placed viewpoints for high-quality synthesis and the quality of the results quickly degrades when the number of captured views is insufficient. To address this problem, we propose a novel NeRF-based framework capable of high-quality view synthesis using only a sparse set of RGB-D images, which can be easily captured using cameras and LiDAR sensors on current consumer devices. First, a geometric proxy of the scene is reconstructed from the captured RGB-D images. Renderings of the reconstructed scene along with precise camera parameters can then be used to pre-train a network. Finally, the network is fine-tuned with a small number of real captured images. We further introduce a patch discriminator to supervise the network under novel views during fine-tuning, as well as a 3D color prior to improve synthesis quality. We demonstrate that our method can generate arbitrary novel views of a 3D scene from as few as 6 RGB-D images. Extensive experiments show the improvements of our method compared with the existing NeRF-based methods, including approaches that also aim to reduce the number of input images.

Neural Radiance Fields (NeRFs) have shown great potential for tasks like novel view synthesis of static 3D scenes. Since NeRFs are trained on a large number of input images, it is not trivial to change their content afterwards. Previous methods to modify NeRFs provide some control but they do not support direct shape deformation which is common for geometry representations like triangle meshes. In this paper, we present a NeRF geometry editing method that first extracts a triangle mesh representation of the geometry inside a NeRF. This mesh can be modified by any 3D modeling tool (we use ARAP mesh deformation). The mesh deformation is then extended into a volume deformation around the shape which establishes a mapping between ray queries to the deformed NeRF and the corresponding queries to the original NeRF. The basic shape editing mechanism is extended towards more powerful and more meaningful editing handles by generating box abstractions of the NeRF shapes which provide an intuitive interface to the user. By additionally assigning semantic labels, we can even identify and combine parts from different objects. We demonstrate the performance and quality of our method in a number of experiments on synthetic data as well as real captured scenes.

Weakly supervised semantic segmentation (WSSS) has gained significant popularity as it relies only on weak labels such as image level annotations rather than the pixel level annotations required by supervised semantic segmentation (SSS) methods. Despite drastically reduced annotation costs, typical feature representations learned from WSSS are only representative of some salient parts of objects and less reliable compared to SSS due to the weak guidance during training. In this paper, we propose a novel Multi-Strategy Contrastive Learning (MuSCLe) framework to obtain enhanced feature representations and improve WSSS performance by exploiting similarity and dissimilarity of contrastive sample pairs at image, region, pixel and object boundary levels. Extensive experiments demonstrate the effectiveness of our method and show that MuSCLe outperforms current state-of-the-art methods on the widely used PASCAL VOC 2012 dataset.

Emotional information plays an important role in various multimedia applications. Movies, as a widely available form of multimedia content, can induce multiple positive emotions and stimulate people's pursuit of a better life. Different from negative emotions, positive emotions are highly correlated and difficult to distinguish in the emotional space. Since different positive emotions are often induced simultaneously by movies, traditional single-target or multi-class methods are not suitable for the classification of movie-induced positive emotions. In this paper, we proposeTransEEG, a model for multi-label positive emotion classification from a viewer's brain activities when watching emotional movies. The key features ofTransEEGinclude (1) explicitly modeling the spatial correlation and temporal dependencies of multi-channel EEG signals using the Transformer structure based model, which effectively addresses long-distance dependencies, (2) exploiting the label-label correlations to guide the discriminative EEG representation learning, for that we design an Inter-Emotion Mask for guiding the Multi-Head Attention to learn the inter-emotion correlations, and (3) constructing an attention score vector from the representation-label correlation matrix to refine emotion-relevant EEG features. To evaluate the ability of our model for multi-label positive emotion classification, we demonstrate our model on a state-of-the-art positive emotion database CPED. Extensive experimental results show that our proposed method achieves superior performance over the competitive approaches.

Benefiting from the intuitiveness and naturalness of sketch interaction, sketch-based video retrieval (SBVR) has received considerable attention in the video retrieval research area. However, most existing SBVR research still lacks the capability of accurate video retrieval with fine-grained scene content. To address this problem, in this paper we investigate a new task, which focuses on retrieving the target video by utilizing a fine-grained storyboard sketch depicting the scene layout and major foreground instances' visual characteristics (e.g., appearance, size, pose, etc.) of video; we call such a task "fine-grained scene-level SBVR". The most challenging issue in this task is how to perform scene-level cross-modal alignment between sketch and video. Our solution consists of two parts. First, we construct a scene-level sketch-video dataset called SketchVideo, in which sketch-video pairs are provided and each pair contains a clip-level storyboard sketch and several keyframe sketches (corresponding to video frames). Second, we propose a novel deep learning architecture called Sketch Query Graph Convolutional Network (SQ-GCN). In SQ-GCN, we first adaptively sample the video frames to improve video encoding efficiency, and then construct appearance and category graphs to jointly model visual and semantic alignment between sketch and video. Experiments show that our fine-grained scene-level SBVR framework with SQ-GCN architecture outperforms the state-of-the-art fine-grained retrieval methods. The SketchVideo dataset and SQ-GCN code are available in the project webpage https://iscas-mmsketch.github.io/FG-SL-SBVR/.

Chronic Glaucoma is an eye disease with progressive optic nerve damage. It is the second leading cause of blindness after cataract and the first leading cause of irreversible blindness. Glaucoma forecast can predict future eye state of a patient by analyzing the historical fundus images, which is helpful for early detection and intervention of potential patients and avoiding the outcome of blindness. In this paper, we propose a GLaucoma forecast transformer based on Irregularly saMpled fundus images named GLIM-Net to predict the probability of developing glaucoma in the future. The main challenge is that the existing fundus images are often sampled at irregular times, making it difficult to accurately capture the subtle progression of glaucoma over time. We therefore introduce two novel modules, namely time positional encoding and time-sensitive MSA (multi-head self-attention) modules, to address this challenge. Unlike many existing works that focus on prediction for an unspecified future time, we also propose an extended model which is further capable of prediction conditioned on a specific future time. The experimental results on the benchmark dataset SIGF show that the accuracy of our method outperforms the state-of-the-art models. In addition, the ablation experiments also confirm the effectiveness of the two modules we propose, which can provide a good reference for the optimization of Transformer models.

With the rapid development of virtual reality (VR) technology, VR headsets, a.k.a. Head-Mounted Displays (HMDs), are widely available, allowing immersive 3D content to be viewed. A natural need for truly immersive VR is to allow bidirectional communication: the user should be able to interact with the virtual world using facial expressions and eye gaze, in addition to traditional means of interaction. The typical application scenario includes VR virtual conferencing and virtual roaming, where ideally users are able to see other users’ expressions and have eye contact with them in the virtual world. In addition, eye gaze also provides a natural means of interaction with virtual objects. Despite significant achievements in recent years for reconstruction of 3D faces from RGB or RGB-D images, it remains a challenge to reliably capture and reconstruct 3D facial expressions including eye gaze when the user is wearing an HMD, because the majority of the face is occluded, especially those areas around the eyes which are essential for recognizing facial expressions and eye gaze. In this paper, we introduce a novel real-time system that is able to capture and reconstruct 3D faces wearing HMDs, and robustly recover eye gaze. We further propose a novel method to map eye gaze directions to the 3D virtual world, which provides a novel and useful interactive mode in VR. We compare our method with state-of-the-art techniques both qualitatively and quantitatively, and demonstrate the effectiveness of our system using live capture.

In this paper, we present a dynamic convolution kernel (DCK) strategy for convolutional neural networks. Using a fully convolutional network with the proposed DCKs, high-quality talking-face video can be generated from multi-modal sources (i.e., unmatched audio and video) in real time, and our trained model is robust to different identities, head postures, and input audios. Our proposed DCKs are specially designed for audio-driven talking face video generation, leading to a simple yet effective end-to-end system. We also provide a theoretical analysis to interpret why DCKs work. Experimental results show that our method can generate high-quality talking-face video with background at 60 fps. Comparison and evaluation between our method and the state-of-the-art methods demonstrate the superiority of our method.

Realistic 3D facial generation based on Neural Radiance Fields (NeRFs) from 2D sketches benefits various applications. Despite the high realism of free-view rendering results of NeRFs, it is tedious and difficult for artists to achieve detailed 3D control and manipulation. Meanwhile, due to its conciseness and expressiveness, sketching has been widely used for 2D facial image generation and editing. Applying sketching to NeRFs is challenging due to the inherent uncertainty for 3D generation with 2D constraints, a significant gap in content richness when generating faces from sparse sketches, and potential inconsistencies for sequential multi-view editing given only 2D sketch inputs. To address these challenges, we present SketchFaceNeRF, a novel sketch-based 3D facial NeRF generation and editing method, to produce free-view photo-realistic images. To solve the challenge of sketch sparsity, we introduce a Sketch Tri-plane Prediction net to first inject the appearance into sketches, thus generating features given reference images to allow color and texture control. Such features are then lifted into compact 3D tri-planes to supplement the absent 3D information, which is important for improving robustness and faithfulness. However, during editing, consistency for unseen or unedited 3D regions is difficult to maintain due to limited spatial hints in sketches. We thus adopt a Mask Fusion module to transform free-view 2D masks (inferred from sketch editing operations) into the tri-plane space as 3D masks, which guide the fusion of the original and sketch-based generated faces to synthesize edited faces. We further design an optimization approach with a novel space loss to improve identity retention and editing faithfulness. Our pipeline enables users to flexibly manipulate faces from different viewpoints in 3D space, easily designing desirable facial models. Extensive experiments validate that our approach is superior to the state-of-the-art 2D sketch-based image generation and editing approaches in realism and faithfulness.

3D shape generation is a fundamental operation in computer graphics. While significant progress has been made, especially with recent deep generative models, it remains a challenge to synthesize high-quality shapes with rich geometric details and complex structures, in a controllable manner. To tackle this, we introduce DSG-Net, a deep neural network that learns a disentangled structured & geometric mesh representation for 3D shapes, where two key aspects of shapes, geometry and structure, are encoded in a synergistic manner to ensure plausibility of the generated shapes, while also being disentangled as much as possible. This supports a range of novel shape generation applications with disentangled control, such as interpolation of structure (geometry) while keeping geometry (structure) unchanged. To achieve this, we simultaneously learn structure and geometry through variational autoencoders (VAEs) in a hierarchical manner for both, with bijective mappings at each level. In this manner, we effectively encode geometry and structure in separate latent spaces, while ensuring their compatibility: the structure is used to guide the geometry and vice versa. At the leaf level, the part geometry is represented using a conditional part VAE, to encode high-quality geometric details, guided by the structure context as the condition. Our method not only supports controllable generation applications, but also produces high-quality synthesized shapes, outperforming state-of-the-art methods.

Accurately estimating the human inner-body under clothing is very important for body measurement, virtual try-on and VR/AR applications. In this article, we propose the first method to allow everyone to easily reconstruct their own 3D inner-body under daily clothing from a self-captured video with the mean reconstruction error of 0.73cm within 15s. This avoids privacy concerns arising from nudity or minimal clothing. Specifically, we propose a novel two-stage framework with a Semantic-guided Undressing Network (SUNet) and an Intra-Inter Transformer Network (IITNet). SUNet learns semantically related body features to alleviate the complexity and uncertainty of directly estimating 3D inner-bodies under clothing. IITNet reconstructs the 3D inner-body model by making full use of intra-frame and inter-frame information, which addresses the misalignment of inconsistent poses in different frames. Experimental results on both public datasets and our collected dataset demonstrate the effectiveness of the proposed method. The code and dataset is available for research purposes at http://cic.tju.edu.cn/faculty/likun/projects/Inner-Body.

Reflectional symmetry is a ubiquitous pattern in nature. Previous works usually solve this problem by voting or sampling, suffering from high computational cost and randomness. In this article, we propose a learning-based approach to intrinsic reflectional symmetry detection. Instead of directly finding symmetric point pairs, we parametrize this self-isometry using a functional map matrix, which can be easily computed given the signs of Laplacian eigenfunctions under the symmetric mapping. Therefore, we manually label the eigenfunction signs for a variety of shapes and train a novel neural network to predict the sign of each eigenfunction under symmetry. Our network aims at learning the global property of functions and consequently converts the problem defined on the manifold to the functional domain. By disentangling the prediction of the matrix into separated bases, our method generalizes well to new shapes and is invariant under perturbation of eigenfunctions. Through extensive experiments, we demonstrate the robustness of our method in challenging cases, including different topology and incomplete shapes with holes. By avoiding random sampling, our learning-based algorithm is over 20 times faster than state-of-the-art methods, and meanwhile, is more robust, achieving higher correspondence accuracy in commonly used metrics.

Coloring line art images based on the colors of reference images is a crucial stage in animation production, which is time-consuming and tedious. This paper proposes a deep architecture to automatically color line art videos with the same color style as the given reference images. Our framework consists of a color transform network and a temporal refinement network based on 3U-net. The color transform network takes the target line art images as well as the line art and color images of the reference images as input and generates corresponding target color images. To cope with the large differences between each target line art image and the reference color images, we propose a distance attention layer that utilizes non-local similarity matching to determine the region correspondences between the target image and the reference images and transforms the local color information from the references to the target. To ensure global color style consistency, we further incorporate Adaptive Instance Normalization (AdaIN) with the transformation parameters obtained from a multiple-layer AdaIN that describes the global color style of the references extracted by an embedder network. The temporal refinement network learns spatiotemporal features through 3D convolutions to ensure the temporal color consistency of the results. Our model can achieve even better coloring results by fine-tuning the parameters with only a small number of samples when dealing with an animation of a new style. To evaluate our method, we build a line art coloring dataset. Experiments show that our method achieves the best performance on line art video coloring compared to the current state-of-the-art methods.

Deformation component analysis is a fundamental problem in geometry processing and shape understanding. Existing approaches mainly extract deformation components in local regions at a similar scale while deformations of real-world objects are usually distributed in a multi-scale manner. In this article, we propose a novel method to exact multiscale deformation components automatically with a stacked attention-based autoencoder. The attention mechanism is designed to learn to softly weight multi-scale deformation components in active deformation regions, and the stacked attention-based autoencoder is learned to represent the deformation components at different scales. Quantitative and qualitative evaluations show that our method outperforms state-of-the-art methods. Furthermore, with the multiscale deformation components extracted by our method, the user can edit shapes in a coarse-to-fine fashion which facilitates effective modeling of new shapes.

Caricature is a type of artistic style of human faces that attracts considerable attention in the entertainment industry. So far a few 3D caricature generation methods exist and all of them require some caricature information (e.g., a caricature sketch or 2D caricature) as input. This kind of input, however, is difficult to provide by non-professional users. In this paper, we propose an end-to-end deep neural network model that generates high-quality 3D caricatures directly from a normal 2D face photo. The most challenging issue for our system is that the source domain of face photos (characterized by normal 2D faces) is significantly different from the target domain of 3D caricatures (characterized by 3D exaggerated face shapes and textures). To address this challenge, we: (1) build a large dataset of 5,343 3D caricature meshes and use it to establish a PCA model in the 3D caricature shape space; (2) reconstruct a normal full 3D head from the input face photo and use its PCA representation in the 3D caricature shape space to establish correspondences between the input photo and 3D caricature shape; and (3) propose a novel character loss and a novel caricature loss based on previous psychological studies on caricatures. Experiments including a novel two-level user study show that our system can generate high-quality 3D caricatures directly from normal face photos.

3D scene stylization aims at generating stylized images of the scene from arbitrary novel views following a given set of style examples, while ensuring consistency when rendered from different views. Directly applying methods for image or video stylization to 3D scenes cannot achieve such consistency. Thanks to recently proposed neural radiance fields (NeRF), we are able to represent a 3D scene in a consistent way. Consistent 3D scene stylization can be effectively achieved by stylizing the corresponding NeRF. However, there is a significant domain gap between style examples which are 2D images and NeRF which is an implicit volumetric representation. To address this problem, we propose a novel mutual learning framework for 3D scene stylization that combines a 2D image stylization network and NeRF to fuse the stylization ability of 2D stylization network with the 3D consistency of NeRF. We first pre-train a standard NeRF of the 3D scene to be stylized and replace its color prediction module with a style network to obtain a stylized NeRF. It is followed by distilling the prior knowledge of spatial consistency from NeRF to the 2D stylization network through an introduced consistency loss. We also introduce a mimic loss to supervise the mutual learning of the NeRF style module and fine-tune the 2D stylization decoder. In order to further make our model handle ambiguities of 2D stylization results, we introduce learnable latent codes that obey the probability distributions conditioned on the style. They are attached to training samples as conditional inputs to better learn the style module in our novel stylized NeRF. Experimental results demonstrate that our method is superior to existing approaches in both visual quality and long-range consistency.

Implicit neural rendering, especially Neural Radiance Field (NeRF), has shown great potential in novel view synthesis of a scene. However, current NeRF-based methods cannot enable users to perform user-controlled shape deformation in the scene. While existing works have proposed some approaches to modify the radiance field according to the user's constraints, the modification is limited to color editing or object translation and rotation. In this paper, we propose a method that allows users to perform controllable shape deformation on the implicit representation of the scene, and synthesizes the novel view images of the edited scene without re-training the network. Specifically, we establish a correspondence between the extracted explicit mesh representation and the implicit neural representation of the target scene. Users can first utilize well-developed mesh-based deformation methods to deform the mesh representation of the scene. Our method then utilizes user edits from the mesh representation to bend the camera rays by introducing a tetrahedra mesh as a proxy, obtaining the rendering results of the edited scene. Extensive experiments demonstrate that our framework can achieve ideal editing results not only on synthetic data, but also on real scenes captured by users.

In this paper, we propose a coarse-to-fine framework to reconstruct a personalized high-fidelity human avatar from a monocular video. To deal with the misalignment problem caused by the changed poses and shapes in different frames, we design a dynamic surface network to recover pose-dependent surface deformations, which help to decouple the shape and texture of the person. To cope with the complexity of textures and generate photo-realistic results, we propose a reference-based neural rendering network and exploit a bottom-up sharpening-guided fine-tuning strategy to obtain detailed textures. Our frame-work also enables photo-realistic novel view/pose syn-thesis and shape editing applications. Experimental re-sults on both the public dataset and our collected dataset demonstrate that our method outperforms the state-of-the-art methods. The code and dataset will be available at http://cic.tju.edu.cn/faculty/likun/projects/HF-Avatar.

Despite the extensive studies on Generative Adversarial Networks (GANs), how to reliably sample high-quality images from their latent spaces remains an under-explored topic. In this paper, we propose a novel GAN latent sampling method by exploring and exploiting the hubness priors of GAN latent distributions. Our key insight is that the high dimensionality of the GAN latent space will inevitably lead to the emergence of hub latents that usually have much larger sampling densities than other latents in the latent space. As a result, these hub latents are better trained and thus contribute more to the synthesis of high-quality images. Unlike the a posterior "cherry-picking", our method is highly efficient as it is an a priori method that identifies high-quality latents before the synthesis of images. Furthermore, we show that the well-known but purely empirical truncation trick is a naive approximation to the central clustering effect of hub latents, which not only uncovers the rationale of the truncation trick, but also indicates the superiority and fundamentality of our method. Extensive experimental results demonstrate the effectiveness of the proposed method. Our code is available at: https://github.com/Byronliang8/HubnessGANSampling.

Mesh saliency, which measures the perceptual importance of different regions on a mesh, benefits a wide range of applications. However, existing mesh saliency models are largely built with hard-coded formulae, which cannot capture true human perception. Some existing techniques utilise indirect measures to capture user perception (e.g., mouse clicks), which can be unreliable. In this work, we collect eye-tracking data for 3D objects seen from different views, and develop an optimisation-based approach to fusing heat-maps captured from individual views to form consistent saliency maps on meshes. To predict mesh saliency on a new shape, we further develop a learning-based approach that regresses local surface characteristics based on a set of input features. Experimental results show that our learning-based method achieves better performance than state-of-the-art methods for unseen shapes. We will make our dataset publicly available.

The advent of deep learning has led to significant progress in monocular human reconstruction. However, existing representations, such as parametric models, voxel grids, meshes and implicit neural representations, have difficulties achieving high-quality results and real-time speed at the same time. In this paper, we propose Fourier Occupancy Field (FOF), a novel, powerful, efficient and flexible 3D geometry representation, for monocular real-time and accurate human reconstruction. A FOF represents a 3D object with a 2D field orthogonal to the view direction where at each 2D position the occupancy field of the object along the view direction is compactly represented with the first few terms of Fourier series, which retains the topology and neighborhood relation in the 2D domain. A FOF can be stored as a multi-channel image, which is compatible with 2D convolutional neural networks and can bridge the gap between 3D geometries and 2D images. A FOF is very flexible and extensible, \eg, parametric models can be easily integrated into a FOF as a prior to generate more robust results. Meshes and our FOF can be easily inter-converted. Based on FOF, we design the first 30+FPS high-fidelity real-time monocular human reconstruction framework. We demonstrate the potential of FOF on both public datasets and real captured data. The code is available for research purposes at http://cic.tju.edu.cn/faculty/likun/projects/FOF.

Abstract Many recent works have reconstructed distinctive 3D face shapes by aggregating shape parameters of the same identity and separating those of different people based on parametric models (e.g. 3D morphable models (3DMMs)). However, despite the high accuracy in the face recognition task using these shape parameters, the visual discrimination of face shapes reconstructed from those parameters remains unsatisfactory. Previous works have not answered the following research question: Do discriminative shape parameters guarantee visual discrimination in represented 3D face shapes? This paper analyses the relationship between shape parameters and reconstructed shape geometry, and proposes a novel shape identity‐aware regularization (SIR) loss for shape parameters, aiming at increasing discriminability in both the shape parameter and shape geometry domains. Moreover, to cope with the lack of training data containing both landmark and identity annotations, we propose a network structure and an associated training strategy to leverage mixed data containing either identity or landmark labels. In addition, since face recognition accuracy does not mean the recognizability of reconstructed face shapes from the shape parameters, we propose the SIR metric to measure the discriminability of face shapes. We compare our method with existing methods in terms of the reconstruction error, visual discriminability, and face recognition accuracy of the shape parameters and SIR metric. Experimental results show that our method outperforms the state‐of‐the‐art methods. The code will be released at https://github.com/a686432/SIR .

Recently, there has been an upsurge of activity in image-based non-photorealistic rendering (NPR), and in particular portrait image stylisation, due to the advent of neural style transfer (NST). However, the state of performance evaluation in this field is poor, especially compared to the norms in the computer vision and machine learning communities. Unfortunately, the task of evaluating image stylisation is thus far not well defined, since it involves subjective, perceptual, and aesthetic aspects. To make progress towards a solution, this paper proposes a new structured, three-level, benchmark dataset for the evaluation of stylised portrait images. Rigorous criteria were used for its construction, and its consistency was validated by user studies. Moreover, a new methodology has been developed for evaluating portrait stylisation algorithms, which makes use of the different benchmark levels as well as annotations provided by user studies regarding the characteristics of the faces. We perform evaluation for a wide variety of image stylisation methods (both portrait-specific and general purpose, and also both traditional NPR approaches and NST) using the new benchmark dataset.

There is a steadily growing range of applications that can benefit from facial reconstruction techniques, leading to an increasing demand for reconstruction of high-quality 3D face models. While it is an important expressive part of the human face, the nose has received less attention than other expressive regions in the face reconstruction literature. When applying existing reconstruction methods to facial images, the reconstructed nose models are often inconsistent with the desired shape and expression. In this paper, we propose a coarse-to-fine 3D nose reconstruction and correction pipeline to build a nose model from a single image, where 3D and 2D nose curve correspondences are adaptively updated and refined. We first correct the reconstruction result coarsely using constraints of 3D-2D sparse landmark correspondences, and then heuristically update a dense 3D-2D curve correspondence based on the coarsely corrected result. A final refinement step is performed to correct the shape based on the updated 3D-2D dense curve constraints. Experimental results show the advantages of our method for 3D nose reconstruction over existing methods.

Spatially localized deformation components are very useful for shape analysis and synthesis in 3D geometry processing. Several methods have recently been developed, with an aim to extract intuitive and interpretable deformation components. However, these techniques suffer from fundamental limitations especially for meshes with noise or large-scale nonlinear deformations, and may not always be able to identify important deformation components. In this paper we propose a mesh-based variational autoencoder architecture that is able to cope with meshes with irregular connectivity and nonlinear deformations, assuming that the analyzed dataset contains meshes with the same vertex connectivity, which is common for deformation analysis. To help localize deformations, we introduce sparse regularization in this framework, along with spectral graph convolutional operations. Through modifying the regularization formulation and allowing dynamic change of sparsity ranges, we improve the visual quality and reconstruction ability of the extracted deformation components. Our system also provides a nonlinear approach to reconstruction of meshes using the extracted basis, which is more effective than the current linear combination approach. As an important application of localized deformation components and a novel approach on its own, we further develop a neural shape editing method, achieving shape editing and deformation component extraction in a unified framework, and ensuring plausibility of the edited shapes. Extensive experiments show that our method outperforms state-of-the-art methods in both qualitative and quantitative evaluations. We also demonstrate the effectiveness of our method for neural shape editing.

Most previous EEG-based emotion recognition methods studied hand-crafted EEG features extracted from different electrodes. In this article, we study the relation among different EEG electrodes and propose a deep learning method to automatically extract the spatial features that characterize the functional relation between EEG signals at different electrodes. Our proposed deep model is calledATtention-basedLSTMwithDomainDiscriminator (ATDD-LSTM), a model based on Long Short-Term Memory (LSTM) for emotion recognition that can characterize nonlinear relations among EEG signals of different electrodes. To achieve state-of-the-art emotion recognition performance, the architecture of ATDD-LSTM has two distinguishing characteristics: (1) By applying the attention mechanism to the feature vectors produced by LSTM, ATDD-LSTM automatically selects suitable EEG channels for emotion recognition, which makes the learned model concentrate on the emotion related channels in response to a given emotion; (2) To minimize the significant feature distribution shift between different sessions and/or subjects, ATDD-LSTM uses a domain discriminator to modify the data representation space and generate domain-invariant features. We evaluate the proposed ATDD-LSTM model on three public EEG emotional databases (DEAP, SEED and CMEED) for emotion recognition. The experimental results demonstrate that our ATDD-LSTM model achieves superior performance on subject-dependent (for the same subject), subject-independent (for different subjects) and cross-session (for the same subject) evaluation.

Convolutional Neural Networks (CNNs) can reveal local variation details and multi-scale spatial context in images via low-to-high stages of feature expression; effective fusion of these raw features is key to Semantic Edge Detection (SED). The methods available in the field generally fuse features across stages in a position-aligned mode, which cannot satisfy the requirements of diverse semantic context in categorizing different pixels. In this paper, we propose a deep framework for SED, the core of which is a new multi-stage feature fusion structure, called All-HiS-In ACA (All-Higher-Stages-In Adaptive Context Aggregation). All-HiS-In ACA can adaptively select semantic context from all higher-stages for detailed features via a cross-stage self-attention paradigm, and thus can obtain fused features with high-resolution details for edge localization and rich semantics for edge categorization. In addition, we develop a non-parametric Inter-layer Complementary Enhancement (ICE) module to supplement clues at each stage with their counterparts in adjacent stages. The ICE-enhanced multi-stage features are then fed into the All-HiS-In ACA module. We also construct an Object-level Semantic Integration (OSI) module to further refine the fused features by enforcing the consistency of the features within the same object. Extensive experiments demonstrate the superior performance of the proposed method over state-of-the-art works.

Sketching is an intuitive and simple way to depict sciences with various object form and appearance characteristics. In the past few years, widely available touchscreen devices have increasingly made sketch-based human-AI co-creation applications popular. One key issue of sketch-oriented interaction is to prepare input sketches efficiently by non-professionals because it is usually difficult and time-consuming to draw an ideal sketch with appropriate outlines and rich details, especially for novice users with no sketching skills. Thus, sketching brings great obstacles for sketch applications in daily life. On the other hand, hand-drawn sketches are scarce and hard to collect. Given the fact that there are several large-scale sketch datasets providing sketch data resources, but they usually have a limited number of objects and categories in sketch, and do not support users to collect new sketch materials according to their personal preferences. In addition, few sketch-related applications support the reuse of existing sketch elements. Thus, knowing how to extract sketches from existing drawings and effectively re-use them in interactive scene sketch composition will provide an elegant way for sketch-based image retrieval (SBIR) applications, which are widely used in various touch screen devices. In this study, we first conduct a study on current SBIR to better understand the main requirements and challenges in sketch-oriented applications. Then we develop the SketchMaker as an interactive sketch extraction and composition system to help users generate scene sketches via reusing object sketches in existing scene sketches with minimal manual intervention. Moreover, we demonstrate how SBIR improves from composited scene sketches to verify the performance of our interactive sketch processing system. We also include a sketch-based video localization task as an alternative application of our sketch composition scheme. Our pilot study shows that our system is effective and efficient, and provides a way to promote practical applications of sketches.

Sketch-based image retrieval (SBIR) is a long-standing research topic in computer vision. Existing methods mainly focus on category-level or instance-level image retrieval. This paper investigates the fine-grained scene-level SBIR problem where a free-hand sketch depicting a scene is used to retrieve desired images. This problem is useful yet challenging mainly because of two entangled facts: 1) achieving an effective representation of the input query data and scene-level images is difficult as it requires to model the information across multiple modalities such as object layout, relative size and visual appearances, and 2) there is a great domain gap between the query sketch input and target images. We present SceneSketcher-v2, a Graph Convolutional Network (GCN) based architecture to address these challenges. SceneSketcher-v2 employs a carefully designed graph convolution network to fuse the multi-modality information in the query sketch and target images and uses a triplet training process and end-to-end training manner to alleviate the domain gap. Extensive experiments demonstrate SceneSketcher-v2 outperforms state-of-the-art scene-level SBIR models with a significant margin.

Multistep tasks, such as block stacking or parts (dis)assembly, are complex for autonomous robotic manipulation. A robotic system for such tasks would need to hierarchically combine motion control at a lower level and symbolic planning at a higher level. Recently, reinforcement learning (RL)-based methods have been shown to handle robotic motion control with better flexibility and generalizability. However, these methods have limited capability to handle such complex tasks involving planning and control with many intermediate steps over a long time horizon. First, current RL systems cannot achieve varied outcomes by planning over intermediate steps (e.g., stacking blocks in different orders). Second, the exploration efficiency of learning multistep tasks is low, especially when rewards are sparse. To address these limitations, we develop a unified hierarchical reinforcement learning framework, named Universal Option Framework (UOF), to enable the agent to learn varied outcomes in multistep tasks. To improve learning efficiency, we train both symbolic planning and kinematic control policies in parallel, aided by two proposed techniques: 1) an auto-adjusting exploration strategy (AAES) at the low level to stabilize the parallel training, and 2) abstract demonstrations at the high level to accelerate convergence. To evaluate its performance, we performed experiments on various multistep block-stacking tasks with blocks of different shapes and combinations and with different degrees of freedom for robot control. The results demonstrate that our method can accomplish multistep manipulation tasks more efficiently and stably, and with significantly less memory consumption.

Sketches, which are simple and concise, have been used in recent deep image synthesis methods to allow intuitive generation and editing of facial images. However, it is nontrivial to extend such methods to video editing due to various challenges, ranging from appropriate manipulation propagation and fusion of multiple editing operations to ensure temporal coherence and visual quality. To address these issues, we propose a novel sketch-based facial video editing framework, in which we represent editing manipulations in latent space and propose specific propagation and fusion modules to generate high-quality video editing results based on StyleGAN3. Specifically, we first design an optimization approach to represent sketch editing manipulations by editing vectors, which are propagated to the whole video sequence using a proper strategy to cope with different editing needs. Specifically, input editing operations are classified into two categories: temporally consistent editing and temporally variant editing. The former (e.g., change of face shape) is applied to the whole video sequence directly, while the latter (e.g., change of facial expression or dynamics) is propagated with the guidance of expression or only affects adjacent frames in a given time window. Since users often perform different editing operations in multiple frames, we further present a region-aware fusion approach to fuse diverse editing effects. Our method supports video editing on facial structure and expression movement by sketch, which cannot be achieved by previous works. Both qualitative and quantitative evaluations show the superior editing ability of our system to existing and alternative solutions.

In this article, we propose a system that can automatically generate immersive and interactive virtual reality (VR) scenes by taking real-world geometric constraints into account. Our system can not only help users avoid real-world obstacles in virtual reality experiences, but also provide context-consistent contents to preserve their sense of presence. To do so, our system first identifies the positions and bounding boxes of scene objects as well as a set of interactive planes from 3D scans. Then context-consistent virtual objects that have similar geometric properties to the real ones can be automatically selected and placed into the virtual scene, based on learned object association relations and layout patterns from large amounts of indoor scene configurations. We regard virtual object replacement as a combinatorial optimization problem, considering both geometric and contextual consistency constraints. Quantitative and qualitative results show that our system can generate plausible interactive virtual scenes that highly resemble real environments, and have the ability to keep the sense of presence for users in their VR experiences.

Realistic speech-driven 3D facial animation is a challenging problem due to the complex relationship between speech and face. In this paper, we propose a deep architecture, called Geometry-guided Dense Perspective Network (GDPnet), to achieve speaker-independent realistic 3D facial animation. The encoder is designed with dense connections to strengthen feature propagation and encourage the re-use of audio features, and the decoder is integrated with an attention mechanism to adaptively recalibrate point-wise feature responses by explicitly modeling interdependencies between different neuron units. We also introduce a non-linear face reconstruction representation as a guidance of latent space to obtain more accurate deformation, which helps solve the geometry-related deformation and is good for generalization across subjects. Huber and HSIC (Hilbert-Schmidt Independence Criterion) constraints are adopted to promote the robustness of our model and to better exploit the non-linear and high-order correlations. Experimental results on the public dataset and real scanned dataset validate the superiority of our proposed GDPnet compared with state-of-the-art model. The code is available for research purposes at http://cic.tju.edu.cn/faculty/likun/projects/GDPnet.

What makes speeches effective has long been a subject for debate, and until today there is broad controversy among public speaking experts about what factors make a speech effective as well as the roles of these factors in speeches. Moreover, there is a lack of quantitative analysis methods to help understand effective speaking strategies. In this paper, we propose E-ffective, a visual analytic system allowing speaking experts and novices to analyze both the role of speech factors and their contribution in effective speeches. From interviews with domain experts and investigating existing literature, we identified important factors to consider in inspirational speeches. We obtained the generated factors from multi-modal data that were then related to effectiveness data. Our system supports rapid understanding of critical factors in inspirational speeches, including the influence of emotions by means of novel visualization methods and interaction. Two novel visualizations include E-spiral (that shows the emotional shifts in speeches in a visually compact way) and E-script (that connects speech content with key speech delivery information). In our evaluation we studied the influence of our system on experts' domain knowledge about speech factors. We further studied the usability of the system by speaking novices and experts on assisting analysis of inspirational speech effectiveness.

3D models are commonly used in computer vision and graphics. With the wider availability of mesh data, an efficient and intrinsic deep learning approach to processing 3D meshes is in great need. Unlike images, 3D meshes have irregular connectivity, requiring careful design to capture relations in the data. To utilize the topology information while staying robust under different triangulations, we propose to encode mesh connectivity using Laplacian spectral analysis, along with mesh feature aggregation blocks (MFABs) that can split the surface domain into local pooling patches and aggregate global information amongst them. We build a mesh hierarchy from fine to coarse using Laplacian spectral clustering, which is flexible under isometric transformations. Inside the MFABs there are pooling layers to collect local information and multi-layer perceptrons to compute vertex features of increasing complexity. To obtain the relationships among different clusters, we introduce a Correlation Net to compute a correlation matrix, which can aggregate the features globally by matrix multiplication with cluster features. Our network architecture is flexible enough to be used on meshes with different numbers of vertices. We conduct several experiments including shape segmentation and classification, and our method outperforms state-of-the-art algorithms for these tasks on the ShapeNet and COSEG datasets.

Synthesizing realistic 3D mesh deformation sequences is a challenging but important task in computer animation. To achieve this, researchers have long been focusing on shape analysis to develop new interpolation and extrapolation techniques. However, such techniques have limited learning capabilities and therefore often produce unrealistic deformation. Although there are already networks defined on individual meshes, deep architectures that operate directly on mesh sequences with temporal information remain unexplored due to the following major barriers: irregular mesh connectivity, rich temporal information, and varied deformation. To address these issues, we utilize convolutional neural networks defined on triangular meshes along with a shape deformation representation to extract useful features, followed by long short-term memory (LSTM) that iteratively processes the features. To fully respect the bidirectional nature of actions, we propose a new share-weight bidirectional scheme to better synthesize deformations. An extensive evaluation shows that our approach outperforms existing methods in sequence generation, both qualitatively and quantitatively.

Cartoon is a common form of art in our daily life and automatic generation of cartoon images from photos is highly desirable. However, state-of-the-art single-style methods can only generate one style of cartoon images from photos and existing multi-style image style transfer methods still struggle to produce high-quality cartoon images due to their highly simplified and abstract nature. In this article, we propose a novel multi-style generative adversarial network (GAN) architecture, called MS-CartoonGAN, which can transform photos into multiple cartoon styles. MS-CartoonGAN uses only unpaired photos and cartoon images of multiple styles for training. To achieve this, we propose to use (1) a hierarchical semantic loss with sparse regularization to retain semantic content and recover flat shading in different abstract levels, (2) a new edge-promoting adversarial loss for producing fine edges, and (3) a style loss to enhance the difference between output cartoon styles and make training process more stable. We also develop a multi-domain architecture, where the generator consists of a shared encoder and multiple decoders for different cartoon styles, along with multiple discriminators for individual styles. By observing that cartoon images drawn by different artists have their unique styles while sharing some common characteristics, our shared network architecture exploits the common characteristics of cartoon styles, achieving better cartoonization and being more efficient than single-style cartoonization. We show that our multi-domain architecture can theoretically guarantee to output desired multiple cartoon styles. Through extensive experiments including a user study, we demonstrate the superiority of the proposed method, outperforming state-of-the-art single-style and multi-style image style transfer methods.

Image colorization is a classic and important topic in computer graphics, where the aim is to add color to a monochromatic input image to produce a colorful result. In this survey, we present the history of colorization research in chronological order and summarize popular algorithms in this field. Early work on colorization mostly focused on developing techniques to improve the colorization quality. In the last few years, researchers have considered more possibilities such as combining colorization with NLP (natural language processing) and focused more on industrial applications. To better control the color, various types of color control are designed, such as providing reference images or color-scribbles. We have created a taxonomy of the colorization methods according to the input type, divided into grayscale, sketch-based and hybrid. The pros and cons are discussed for each algorithm, and they are compared according to their main characteristics. Finally, we discuss how deep learning, and in particular Generative Adversarial Networks (GANs), has changed this field.

People with low vision experience reduced mobility that affects their physical and mental wellbeing. With augmented reality (AR) glasses, there are new opportunities to provide visual and auditory information that can improve mobility for this vulnerable group. Current research into AR-based mobility aids has focused mainly on the technical aspects, and less emphasis has been placed on understanding the usability and suitability of these aids in people with various levels of visual impairment. In this paper, we present the results of qualitative interviews with 18 participants using HoloLens v1 and eight prototype augmentations to understand how these enhancements are perceived by people with low vision and how these aids should be adjusted to suit their needs. Our results suggested that participants with moderate vision loss could potentially perceive the most benefit from glasses and underlined the importance of extensive customizability to accommodate the needs of a highly varied low vision population.

Learning computational models of image aesthetics can have a substantial impact on visual art and graphic design. Although automatic image aesthetics assessment is a challenging topic by its subjective nature, psychological studies have confirmed a strong correlation between image layouts and perceived image quality. While previous state-of-the-art methods attempt to learn holistic information using deep Convolutional Neural Networks (CNNs), our approach is motivated by the fact that Graph Convolutional Network (GCN) architecture is conceivably more suited for modeling complex relations among image regions than vanilla convolutional layers. Specifically, we present a Hierarchical Layout-Aware Graph Convolutional Network (HLA-GCN) to capture layout information. It is a dedicated double-subnet neural network consisting of two LA-GCN modules. The first LA-GCN module constructs an aesthetics-related graph in the coordinate space and performs reasoning over spatial nodes. The second LA-GCN module performs graph reasoning after aggregating significant regions in a latent space. The model output is a hierarchical representation with layout-aware features from both spatial and aggregated nodes for unified aesthetics assessment. Extensive evaluations show that our proposed model outperforms the state-of-the-art on the AVA and AADB datasets across three different tasks. The code is available at http://github.com/days1011/HLAGCN.

Person image synthesis, e.g., pose transfer, is a challenging problem due to large variation and occlusion. Existing methods have difficulties predicting reasonable invisible regions and fail to decouple the shape and style of clothing, which limits their applications on person image editing. In this paper, we propose PISE, a novel two-stage generative model for Person Image Synthesis and Editing, which is able to generate realistic person images with desired poses, textures, or semantic layouts. For human pose transfer, we first synthesize a human parsing map aligned with the target pose to represent the shape of clothing by a parsing generator, and then generate the final image by an image generator. To decouple the shape and style of clothing, we propose joint global and local per-region encoding and normalization to predict the reasonable style of clothing for invisible regions. We also propose spatial-aware normalization to retain the spatial context relationship in the source image. The results of qualitative and quantitative experiments demonstrate the superiority of our model on human pose transfer. Besides, the results of texture transfer and region editing show that our model can be applied to person image editing. The code is available for research purposes at https://github.com/Zhangjinso/PISE.

Most existing style transfer methods follow the assumption that styles can be represented with global statistics (e.g., Gram matrices or covariance matrices), and thus address the problem by forcing the output and style images to have similar global statistics. An alternative is the assumption of local style patterns, where algorithms are designed to swap similar local features of content and style images. However, the limitation of these existing methods is that they neglect the semantic structure of the content image which may lead to corrupted content structure in the output. In this paper, we make a new assumption that image features from the same semantic region form a manifold and an image with multiple semantic regions follows a multi-manifold distribution. Based on this assumption, the style transfer problem is formulated as aligning two multi-manifold distributions and a Manifold Alignment based Style Transfer (MAST) framework is proposed. The proposed frame-work allows semantically similar regions between the output and the style image share similar style patterns. Moreover, the proposed manifold alignment method is flexible to allow user editing or using semantic segmentation maps as guidance for style transfer. To allow the method to be applicable to photorealistic style transfer, we propose a new adaptive weight skip connection network structure to preserve the content details. Extensive experiments verify the effectiveness of the proposed framework for both artistic and photorealistic style transfer. Code is available at https://github.com/NJUHuoJing/MAST.

In this work, we tackle the problem of single image-based 3D shape retrieval (IBSR), where we seek to find the most matched shape of a given single 2D image from a shape repository. Most of the existing works learn to embed 2D images and 3D shapes into a common feature space and perform metric learning using a triplet loss. Inspired by the great success in recent contrastive learning works on self-supervised representation learning, we propose a novel IBSR pipeline leveraging contrastive learning. We note that adopting such cross-modal contrastive learning between 2D images and 3D shapes into IBSR tasks is non-trivial and challenging: contrastive learning requires very strong data augmentation in constructed positive pairs to learn the feature invariance, whereas traditional metric learning works do not have this requirement. Moreover, object shape and appearance are entangled in 2D query images, thus making the learning task more difficult than contrasting single-modal data. To mitigate the challenges, we propose to use multi-view grayscale rendered images from the 3D shapes as a shape representation. We then introduce a strong data augmentation technique based on color transfer, which can significantly but naturally change the appearance of the query image, effectively satisfying the need for contrastive learning. Finally, we propose to incorporate a novel category-level contrastive loss that helps distinguish similar objects from different categories, in addition to classic instance-level contrastive loss. Our experiments demonstrate that our approach achieves the best performance on all the three popular IBSR benchmarks, including Pix3D, Stanford Cars, and Comp Cars, outperforming the previous state-of-the-art from 4% - 15% on retrieval accuracy.

Benefiting from the excellent performance of Siamese-based trackers, huge progress on 2D visual tracking has been achieved. However, 3D visual tracking is still under-explored. Inspired by the idea of Hough voting in 3D object detection, in this paper, we propose a Multi-level Voting Siamese Network (MLVSNet) for 3D visual tracking from outdoor point cloud sequences. To deal with sparsity in outdoor 3D point clouds, we propose to perform Hough voting on multi-level features to get more vote centers and retain more useful information, instead of voting only on the fi-nal level feature as in previous methods. We also design an efficient and lightweight Target-Guided Attention (TGA) module to transfer the target information and highlight the target points in the search area. Moreover, we propose a Vote-cluster Feature Enhancement (VFE) module to exploit the relationships between different vote clusters. Extensive experiments on the 3D tracking benchmark of KITTI dataset demonstrate that our MLVSNet outperforms state-of-the-art methods with significant margins. Code will be available at https://github.com/CodeWZT/MLVSNet.

Hough voting, as has been demonstrated in VoteNet, is effective for 3D object detection, where voting is a key step. In this paper, we propose a novel VoteNet-based 3D detector with vote enhancement to improve the detection accuracy in cluttered indoor scenes. It addresses the limitations of current voting schemes, i.e., votes from neighboring objects and background have significant negative impacts. Before voting, we replace the classic MLP with the proposed Attentive MLP (AMLP) in the backbone network to get better feature description of seed points. During voting, we design a new vote attraction loss (VALoss) to enforce vote centers to locate closely and compactly to the corresponding object centers. After voting, we then devise a vote weighting module to integrate the foreground/background prediction into the vote aggregation process to enhance the capability of the original VoteNet to handle noise from background voting. The three proposed strategies all contribute to more effective voting and improved performance, resulting in a novel 3D object detector, termed VENet. Experiments show that our method outperforms state-of-the-art methods on benchmark datasets. Ablation studies demonstrate the effectiveness of the proposed components.

We present a practical backend for stereo visual SLAM which can simultaneously discover individual rigid bodies and compute their motions in dynamic environments. While recent factor graph based state optimization algorithms have shown their ability to robustly solve SLAM problems by treating dynamic objects as outliers, their dynamic motions are rarely considered. In this paper, we exploit the consensus of 3D motions for landmarks extracted from the same rigid body for clustering, and to identify static and dynamic objects in a unified manner. Specifically, our algorithm builds a noise-aware motion affinity matrix from landmarks, and uses agglomerative clustering to distinguish rigid bodies. Using decoupled factor graph optimization to revise their shapes and trajectories, we obtain an iterative scheme to update both cluster assignments and motion estimation reciprocally. Evaluations on both synthetic scenes and KITTI demonstrate the capability of our approach, and further experiments considering online efficiency also show the effectiveness of our method for simultaneously tracking ego-motion and multiple objects.

Despite significant effort and notable success of neural style transfer, it remains challenging for highly abstract styles, in particular line drawings. In this paper, we propose APDrawingGAN++, a generative adversarial network (GAN) for transforming face photos to artistic portrait drawings (APDrawings), which addresses substantial challenges including highly abstract style, different drawing techniques for different facial features, and high perceptual sensitivity to artifacts. To address these, we propose a composite GAN architecture that consists of local networks (to learn effective representations for specific facial features) and a global network (to capture the overall content). We provide a theoretical explanation for the necessity of this composite GAN structure by proving that any GAN with a single generator cannot generate artistic styles like APDrawings. We further introduce a classification-and-synthesis approach for lips and hair where different drawing styles are used by artists, which applies suitable styles for a given input. To capture the highly abstract art form inherent in APDrawings, we address two challenging operations-(1) coping with lines with small misalignments while penalizing large discrepancy and (2) generating more continuous lines-by introducing two novel loss terms: one is a novel distance transform loss with nonlinear mapping and the other is a novel line continuity loss, both of which improve the line quality. We also develop dedicated data augmentation and pre-training to further improve results. Extensive experiments, including a user study, show that our method outperforms state-of-the-art methods, both qualitatively and quantitatively.

Over-segmenting a video into supervoxels has strong potential to reduce the complexity of downstream computer vision applications. Content-sensitive supervoxels (CSSs) are typically smaller in content-dense regions (i.e., with high variation of appearance and/or motion) and larger in content-sparse regions. In this paper, we propose to compute feature-aware CSSs (FCSSs) that are regularly shaped 3D primitive volumes well aligned with local object/region/motion boundaries in video. To compute FCSSs, we map a video to a 3D manifold embedded in a combined color and spatiotemporal space, in which the volume elements of video manifold give a good measure of the video content density. Then any uniform tessellation on video manifold can induce CSS in the video. Our idea is that among all possible uniform tessellations on the video manifold, FCSS finds one whose cell boundaries well align with local video boundaries. To achieve this goal, we propose a novel restricted centroidal Voronoi tessellation method that simultaneously minimizes the tessellation energy (leading to uniform cells in the tessellation) and maximizes the average boundary distance (leading to good local feature alignment). Theoretically our method has an optimal competitive ratio O(1), and its time and space complexities are O(NK) and O(N+K) for computing K supervoxels in an N-voxel video. We also present a simple extension of FCSS to streaming FCSS for processing long videos that cannot be loaded into main memory at once. We evaluate FCSS, streaming FCSS and ten representative supervoxel methods on four video datasets and two novel video applications. The results show that our method simultaneously achieves state-of-the-art performance with respect to various evaluation criteria.

Micro-expressions are the rapid movements of facial muscles that can be used to reveal concealed emotions. Recognizing them from video clips has a wide range of applications and receives increasing attention recently. Among existing methods, the main directional mean optical-flow (MDMO) feature achieves state-of-the-art performance for recognizing spontaneous micro-expressions. For a video clip, the MDMO feature is computed by averaging a set of atomic features frame-by-frame. Despite its simplicity, the average operation in MDMO can easily lose the underlying manifold structure inherent in the feature space. In this paper we propose a sparse MDMO feature that learns an effective dictionary from a micro-expression video dataset. In particular, a new distance metric is proposed based on the sparsity of sample points in the MDMO feature space, which can efficiently reveal the underlying manifold structure. The proposed sparse MDMO feature is obtained by incorporating this new metric into the classic graph regularized sparse coding (GraphSC) scheme. We evaluate sparse MDMO and four representative features (LBP-TOP, STCLQP, MDMO and FDM) on three spontaneous micro-expression datasets (SMIC, CASME and CASME II). The results show that sparse MDMO outperforms these representative features.

Given an input face photo, the goal of caricature generation is to produce stylized, exaggerated caricatures that share the same identity as the photo. It requires simultaneous style transfer and shape exaggeration with rich diversity, and meanwhile preserving the identity of the input. To address this challenging problem, we propose a novel framework called Multi-Warping GAN (MW-GAN), including a style network and a geometric network that are designed to conduct style transfer and geometric exaggeration respectively. We bridge the gap between the style/landmark space and their corresponding latent code spaces by a dual way design, so as to generate caricatures with arbitrary styles and geometric exaggeration, which can be specified either through random sampling of latent code or from a given caricature sample. Besides, we apply identity preserving loss to both image space and landmark space, leading to a great improvement in quality of generated caricatures. Experiments show that caricatures generated by MW-GAN have better quality than existing methods.

3D human reconstruction from a single image is a challenging problem. Existing methods have difficulties to infer 3D clothed human models with consistent topologies for various poses. In this paper, we propose an efficient and effective method using a hierarchical graph transformation network. To deal with large deformations and avoid distorted geometries, rather than using Euclidean coordinates directly, 3D human shapes are represented by a vertex-based deformation representation that effectively encodes the deformation and copes well with large deformations. To infer a 3D human mesh consistent with the input real image, we also use a perspective projection layer to incorporate perceptual image features into the deformation representation. Our model is easy to train and fast to converge with short test time. Besides, we present the$D^{2}Human$(Dynamic Detailed Human) dataset, including variously posed 3D human meshes with consistent topologies and rich geometry details, together with the captured color images and SMPL models, which is useful for training and evaluation of deep frameworks, particularly for graph neural networks. Experimental results demonstrate that our method achieves more plausible and complete 3D human reconstruction from a single image, compared with several state-of-the-art methods. The code and dataset are available for research purposes athttp://cic.tju.edu.cn/faculty/likun/projects/MGTnet.

Deep learning (DL) is inherently subject to the requirement of a large amount of well-labeled data, which is expensive and time-consuming to obtain manually. In order to broaden the reach of DL, leveraging free web data becomes an attractive strategy to alleviate the issue of data scarcity. However, directly utilizing collected web data to train a deep model is ineffective because of the mixed noisy data. To address such problems, we develop a novel bidirectional self-paced learning (BiSPL) framework which reduces the effect of noise by learning from web data in a meaningful order. Technically, the BiSPL framework consists of two essential steps. Relying on distances defined between web samples and labeled source samples, first, the web samples with short distances are sampled and combined to form a new training set. Second, based on the new training set, both easy and hard samples are initially employed to train deep models for higher stability, and hard samples are gradually dropped to reduce the noise as the training progresses. By iteratively alternating such steps, deep models converge to a better solution. We mainly focus on the fine-grained visual classification (FGVC) tasks because their corresponding datasets are generally small and therefore face a more significant data scarcity problem. Experiments conducted on six public FGVC tasks demonstrate that our proposed method outperforms the state-of-the-art approaches. Especially, BiSPL suffices to achieve the highest stable performance when the scale of the well-labeled training set decreases dramatically.

With the popularity of social media, an increasing number of people are accustomed to expressing their feelings and emotions online using images and videos. An emotion-based image retrieval (EBIR) system is useful for obtaining visual contents with desired emotions from a massive repository. Existing EBIR methods mainly focus on modeling the global characteristics of visual content without considering the crucial role of informative regions of interest in conveying emotions. Further, they ignore the hierarchical relationships between coarse polarities and fine categories of emotions. In this paper, we design an attention-aware polarity-sensitive embedding (APSE) network to address these issues. First, we develop a hierarchical attention mechanism to automatically discover and model the informative regions of interest. Specifically, both polarity- and emotion-specific attended representations are aggregated for discriminative feature embedding. Second, we propose a generated emotion-pair (GEP) loss to simultaneously consider the inter- and intra-polarity relationships of the emotion labels. Moreover, we adaptively generate negative examples of different hard levels in the feature space guided by the attention module to further improve the performance of feature embedding. Extensive experiments on four popular benchmark datasets demonstrate that the proposed APSE method outperforms the state-of-the-art EBIR approaches by a large margin.

Recent facial image synthesis methods have been mainly based on conditional generative models. Sketch-based conditions can effectively describe the geometry of faces, including the contours of facial components, hair structures, as well as salient edges (e.g., wrinkles) on face surfaces but lack effective control of appearance, which is influenced by color, material, lighting condition, etc. To have more control of generated results, one possible approach is to apply existing disentangling works to disentangle face images into geometry and appearance representations. However, existing disentangling methods are not optimized for human face editing, and cannot achieve fine control of facial details such as wrinkles. To address this issue, we propose DeepFaceEditing, a structured disentanglement framework specifically designed for face images to support face generation and editing with disentangled control of geometry and appearance. We adopt a local-to-global approach to incorporate the face domain knowledge: local component images are decomposed into geometry and appearance representations, which are fused consistently using a global fusion module to improve generation quality. We exploit sketches to assist in extracting a better geometry representation, which also supports intuitive geometry editing via sketching. The resulting method can either extract the geometry and appearance representations from face images, or directly extract the geometry representation from face sketches. Such representations allow users to easily edit and synthesize face images, with decoupled control of their geometry and appearance. Both qualitative and quantitative evaluations show the superior detail and appearance control abilities of our method compared to state-of-the-art methods.

We introduce TM-NET, a novel deep generative model for synthesizing textured meshes in a part-aware manner. Once trained, the network can generate novel textured meshes from scratch or predict textures for a given 3D mesh, without image guidance. Plausible and diverse textures can be generated for the same mesh part, while texture compatibility between parts in the same shape is achieved via conditional generation. Specifically, our method produces texture maps for individual shape parts, each as a deformable box, leading to a natural UV map with limited distortion. The network separately embeds part geometry (via a PartVAE) and part texture (via a TextureVAE) into their respective latent spaces, so as to facilitate learning texture probability distributions conditioned on geometry. We introduce a conditional autoregressive model for texture generation, which can be conditioned on both part geometry and textures already generated for other parts to achieve texture compatibility. To produce high-frequency texture details, our TextureVAE operates in a high-dimensional latent space via dictionary-based vector quantization. We also exploit transparencies in the texture as an effective means to model complex shape structures including topological details. Extensive experiments demonstrate the plausibility, quality, and diversity of the textures and geometries generated by our network, while avoiding inconsistency issues that are common to novel view synthesis methods.

In this paper, we present a new computational pipeline for designing and fabricating 4D garments as knitwear that considers comfort during body movement. This is achieved by careful control of elasticity distribution to reduce uncomfortable pressure and unwanted sliding caused by body motion. We exploit the ability to knit patterns in different elastic levels by single-jersey jacquard (SJJ) with two yarns. We design the distribution of elasticity for a garment by physics-based computation, the optimized elasticity on the garment is then converted into instructions for a digital knitting machine by two algorithms proposed in this paper. Specifically, a graph-based algorithm is proposed to generate knittable stitch meshes that can accurately capture the 3D shape of a garment, and a tiling algorithm is employed to assign SJJ patterns on the stitch mesh to realize the designed distribution of elasticity. The effectiveness of our approach is verified on simulation results and on specimens physically fabricated by knitting machines.

Example-based mesh deformation methods are powerful tools for realistic shape editing. However, existing techniques typically combine all the example deformation modes, which can lead to overfitting, i.e., using an overly complicated model to explain the user-specified deformation. This leads to implausible or unstable deformation results, including unexpected global changes outside the region of interest. To address this fundamental limitation, we propose a sparse blending method that automatically selects a smaller number of deformation modes to compactly describe the desired deformation. This along with a suitably chosen deformation basis including spatially localized deformation modes leads to significant advantages, including more meaningful, reliable, and efficient deformations because fewer and localized deformation modes are applied. To cope with large rotations, we develop a simple but effective representation based on polar decomposition of deformation gradients, which resolves the ambiguity of large global rotations using an as-consistent-as-possible global optimization. This simple representation has a closed form solution for derivatives, making it efficient for our sparse localized representation and thus ensuring interactive performance. Experimental results show that our method outperforms state-of-the-art data-driven mesh deformation methods, for both quality of results and efficiency.

In geometry processing, symmetry is a universal type of high-level structural information of 3D models and benefits many geometry processing tasks including shape segmentation, alignment, matching, and completion. Thus it is an important problem to analyze various symmetry forms of 3D shapes. Planar reflective symmetry is the most fundamental one. Traditional methods based on spatial sampling can be time-consuming and may not be able to identify all the symmetry planes. In this article, we present a novel learning framework to automatically discover global planar reflective symmetry of a 3D shape. Our framework trains an unsupervised 3D convolutional neural network to extract global model features and then outputs possible global symmetry parameters, where input shapes are represented using voxels. We introduce a dedicated symmetry distance loss along with a regularization loss to avoid generating duplicated symmetry planes. Our network can also identify generalized cylinders by predicting their rotation axes. We further provide a method to remove invalid and duplicated planes and axes. We demonstrate that our method is able to produce reliable and accurate results. Our neural network based method is hundreds of times faster than the state-of-the-art methods, which are based on sampling. Our method is also robust even with noisy or incomplete input surfaces.

Small object arrangement is very important for creating detailed and realistic 3D indoor scenes. In this article, we present an interactive framework based on active learning to help users create customized arrangements for small objects according to their preferences. To achieve this with minimal user effort, we first learn the prior knowledge about small object arrangement from a 3D indoor scene dataset through a probability mining method, which forms the initial guidance for arranging small objects. Then, users are able to express their preferences on a few small object categories, which are automatically propagated to all the other categories via a novel active learning approach. In the propagation process, we introduce a novel metric to obtain the propagation weights, which measures the degree of interchangeability between two small object categories, and is calculated based on a spatial embedding model learned from the small object neighborhood information extracted from the 3D indoor scene dataset. Experiments show that our framework is able to help users effectively create customized small object arrangements with little effort.

This article proposes an approach to content-preserving image stitching with regular boundary constraints, which aims to stitch multiple images to generate a panoramic image with piecewise rectangular boundaries. Existing methods treat image stitching and rectangling as two separate steps, which may result in suboptimal results as the stitching process is not aware of the further warping needs for rectangling. We address these limitations by formulating image stitching with regular boundaries in a unified optimization framework. Starting from the initial stitching result produced by traditional warping-based optimization, we obtain the irregular boundary from the warped meshes by polygon Boolean operations which robustly handle arbitrary mesh compositions. By analyzing the irregular boundary, we construct a piecewise rectangular boundary. Based on this, we further incorporate line and regular boundary preservation constraints into the image stitching framework, and conduct iterative optimizations to obtain an optimal piecewise rectangular boundary. Thus we can make the boundary of the stitching result as close as possible to a rectangle, while reducing unwanted distortions. We further extend our method to video stitching, by integrating the temporal coherence into the optimization. Experiments show that our method efficiently produces visually pleasing panoramas with regular boundaries and unnoticeable distortions.

In this paper, we address the 3D object detection task by capturing multi-level contextual information with the self-attention mechanism and multi-scale feature fusion. Most existing 3D object detection methods recognize objects individually, without giving any consideration on contextual information between these objects. Comparatively, we propose Multi-Level Context VoteNet (MLCVNet) to recognize 3D objects correlatively, building on the state-of-the-art VoteNet. We introduce three context modules into the voting and classifying stages of VoteNet to encode contextual information at different levels. Specifically, a Patch-to-Patch Context (PPC) module is employed to capture contextual information between the point patches, before voting for their corresponding object centroid points. Subsequently, an Object-to-Object Context (OOC) module is incorporated before the proposal and classification stage, to capture the contextual information between object candidates. Finally, a Global Scene Context (GSC) module is designed to learn the global scene context. We demonstrate these by capturing contextual information at patch, object and scene levels. Our method is an effective way to promote detection accuracy, achieving new state-of-the-art detection performance on challenging 3D object detection datasets, i.e., SUN RGBD and ScanNet. We also release our code at https://github.com/NUAAXQ/MLCVNet.

Portrait drawing is a common form of art with high abstraction and expressiveness. Due to its unique characteristics, existing methods achieve decent results only with paired training data, which is costly and time-consuming to obtain.In this paper, we address the problem of automatic transfer from face photos to portrait drawings with unpaired training data. We observe that due to the significant imbalance of information richness between photos and drawings, existing unpaired transfer methods such as CycleGAN tends to embed invisible reconstruction information indiscriminately in the whole drawings, leading to important facial features partially missing in drawings. To address this problem, we propose a novel asymmetric cycle mapping that enforces the reconstruction information to be visible (by a truncation loss) and only embedded in selective facial regions (by a relaxed forward cycle-consistency loss). Along with localized discriminators for the eyes, nose and lips, our method well preserves all important facial features in the generated portrait drawings. By introducing a style classifier and taking the style vector into account, our method can learn to generate portrait drawings in multiple styles using a single network. Extensive experiments show that our model outperforms state-of-the-art methods.

Estimating correspondence between two shapes continues to be a challenging problem in geometry processing. Most current methods assume deformation to be near-isometric, however this is often not the case. For this paper, a collection of shapes of different animals has been curated, where parts of the animals (e.g., mouths, tails & ears) correspond yet are naturally non-isometric. Ground-truth correspondences were established by asking three specialists to independently label corresponding points on each of the models with respect to a previously labelled reference model. We employ an algorithmic strategy to select a single point for each correspondence that is representative of the proposed labels. A novel technique that characterises the sparsity and distribution of correspondences is employed to measure the performance of ten shape correspondence methods.

Abstract Image‐to‐image translation has been widely studied. Since real‐world images can often be described by multiple attributes, it is useful to manipulate them at the same time. However, most methods focus on transforming between two domains, and when they chain multiple single attribute transform networks together, the results are affected by the order of chaining, and the performance drops with the out‐of‐domain issue for intermediate results. Existing multi‐domain transfer methods mostly manipulate multiple attributes by adding a list of attribute labels to the network feature, but they also suffer from interference of different attributes, and perform worse when multiple attributes are manipulated. We propose a novel approach to multi‐attribute image‐to‐image translation using several parallel latent transform networks, where multiple attributes are manipulated in parallel and simultaneously, which eliminates both issues. To avoid the interference of different attributes, we introduce a novel soft independence constraint for the changes caused by different attributes. Extensive experiments show that our method outperforms state‐of‐the‐art methods.

Humans have the ability to perceive kinetic depth effects , i.e., to perceived 3D shapes from 2D projections of rotating 3D objects. This process is based on a variety of visual cues such as lighting and shading effects. However, when such cues are weak or missing, perception can become faulty, as demonstrated by the famous silhouette illusion example of the spinning dancer . Inspired by this, we establish objective and subjective evaluation models of rotated 3D objects by taking their projected 2D images as input. We investigate five different cues: ambient luminance, shading, rotation speed, perspective, and color difference between the objects and background. In the objective evaluation model, we first apply 3D reconstruction algorithms to obtain an objective reconstruction quality metric, and then use quadratic stepwise regression analysis to determine weights of depth cues to represent the reconstruction quality. In the subjective evaluation model, we use a comprehensive user study to reveal correlations with reaction time and accuracy, rotation speed, and perspective. The two evaluation models are generally consistent, and potentially of benefit to inter-disciplinary research into visual perception and 3D reconstruction.

Researchers have achieved great success in dealing with 2D images using deep learning. In recent years, 3D computer vision and geometry deep learning have gained ever more attention. Many advanced techniques for 3D shapes have been proposed for different applications. Unlike 2D images, which can be uniformly represented by a regular grid of pixels, 3D shapes have various representations, such as depth images, multi-view images, voxels, point clouds, meshes, implicit surfaces, etc. The performance achieved in different applications largely depends on the representation used, and there is no unique representation that works well for all applications. Therefore, in this survey, we review recent developments in deep learning for 3D geometry from a representation perspective, summarizing the advantages and disadvantages of different representations for different applications. We also present existing datasets in these representations and further discuss future research directions.

Image retargeting techniques adjust images into different sizes and have attracted much attention recently. Objective quality assessment (OQA) of image retargeting results is often desired to automatically select the best results. Existing OQA methods train a model using some benchmarks (e.g., RetargetMe), in which subjective scores evaluated by users are provided. Observing that it is challenging even for human subjects to give consistent scores for retargeting results of different source images (diff-source-results), in this paper we propose a learning-based OQA method that trains a General Regression Neural Network (GRNN) model based on relative scores-which preserve the ranking-of retargeting results of the same source image (same-source-results). In particular, we develop a novel training scheme with provable convergence that learns a common base scalar for same-source-results. With this source specific offset, our computed scores not only preserve the ranking of subjective scores for same-source-results, but also provide a reference to compare the diff-source-results. We train and evaluate our GRNN model using human preference data collected in RetargetMe. We further introduce a subjective benchmark to evaluate the generalizability of different OQA methods. Experimental results demonstrate that our method outperforms ten representative OQA methods in ranking prediction and has better generalizability to different datasets.

This paper addresses the challenge of 3D motion recovery by exploiting the spatio-temporal correlations of corrupted 3D skeleton sequences. We propose a new 3D motion recovery method using spatio-temporal reconstruction, which uses joint low-rank and sparse priors to exploit temporal correlation and an isometric constraint for spatial correlation. The proposed model is formulated as a constrained optimization problem, which is efficiently solved by the augmented Lagrangian method with a Gauss-Newton solver for the subproblem of isometric optimization. The experimental results on the CMU motion capture dataset, Edinburgh dataset, and two Kinect datasets demonstrate that the proposed approach achieves better motion recovery than the state-of-the-art methods. The proposed method is applicable to Kinect-like skeleton tracking devices and pose estimation methods that cannot provide accurate estimation of complex motions, especially in the presence of occlusion.

State-of-the-art neural style transfer methods have demonstrated amazing results by training feed-forward convolutional neural networks or using an iterative optimization strategy. The image representation used in these methods, which contains two components: style representation and content representation, is typically based on high-level features extracted from pretrained classification networks. Because the classification networks are originally designed for object recognition, the extracted features often focus on the central object and neglect other details. As a result, the style textures tend to scatter over the stylized outputs and disrupt the content structures. To address this issue, we present a novel image stylization method that involves an additional structure representation. Our structure representation, which considers two factors: i) the global structure represented by the depth map and ii) the local structure details represented by the image edges, effectively reflects the spatial distribution of all the components in an image as well as the structure of dominant objects respectively. Experimental results demonstrate that our method achieves an impressive visual effectiveness, which is particularly significant when processing images sensitive to structure distortion, e.g. images containing multiple objects potentially at different depths, or dominant objects with clear structures.

Mesh color edit propagation aims to propagate the color from a few color strokes to the whole mesh, which is useful for mesh colorization, color enhancement and color editing, etc. Compared with image edit propagation, luminance information is not available for 3D mesh data, so the color edit propagation is more difficult on 3D meshes than images, with far less research carried out. This paper proposes a novel solution based on sparse graph regularization. Firstly, a few color strokes are interactively drawn by the user, and then the color will be propagated to the whole mesh by minimizing a sparse graph regularized nonlinear energy function. The proposed method effectively measures geometric similarity over shapes by using a set of complementary multiscale feature descriptors, and effectively controls color bleeding via a sparse ℓ1 optimization rather than quadratic minimization used in existing work. The proposed framework can be applied for the task of interactive mesh colorization, mesh color enhancement and mesh color editing. Extensive qualitative and quantitative experiments show that the proposed method outperforms the state-of-the-art methods.

Human pose transfer, which aims at transferring the appearance of a given person to a target pose, is very challenging and important in many applications. Previous work ignores the guidance of pose features or only uses local attention mechanism, leading to implausible and blurry results. We propose a new human pose transfer method using a generative adversarial network (GAN) with simplified cascaded blocks. In each block, we propose a pose-guided non-local attention (PoNA) mechanism with a long-range dependency scheme to select more important regions of image features to transfer. We also design pre-posed image-guided pose feature update and post-posed pose-guided image feature update to better utilize the pose and image features. Our network is simple, stable, and easy to train. Quantitative and qualitative results on Market-1501 and DeepFashion datasets show the efficacy and efficiency of our model. Compared with state-of-the-art methods, our model generates sharper and more realistic images with rich details, while having fewer parameters and faster speed. Furthermore, our generated images can help to alleviate data insufficiency for person re-identification.

This paper proposes a new method for simultaneous 3D reconstruction and semantic segmentation for indoor scenes. Unlike existing methods that require recording a video using a color camera and/or a depth camera, our method only needs a small number of (e.g., 3~5) color images from uncalibrated sparse views, which significantly simplifies data acquisition and broadens applicable scenarios. To achieve promising 3D reconstruction from sparse views with limited overlap, our method first recovers the depth map and semantic information for each view, and then fuses the depth maps into a 3D scene. To this end, we design an iterative deep architecture, named IterNet, to estimate the depth map and semantic segmentation alternately. To obtain accurate alignment between views with limited overlap, we further propose a joint global and local registration method to reconstruct a 3D scene with semantic information. We also make available a new indoor synthetic dataset, containing photorealistic high-resolution RGB images, accurate depth maps and pixel-level semantic labels for thousands of complex layouts. Experimental results on public datasets and our dataset demonstrate that our method achieves more accurate depth estimation, smaller semantic segmentation errors, and better 3D reconstruction results over state-of-the-art methods.

Automatic assessment of sentiment from visual content has gained considerable attention with the increasing tendency of expressing opinions online. In this paper, we solve the problem of visual sentiment analysis, which is challenging due to the high-level abstraction in the recognition process. Existing methods based on convolutional neural networks learn sentiment representations from the holistic image, despite the fact that different image regions can have different influence on the evoked sentiment. In this paper, we introduce a weakly supervised coupled convolutional network (WSCNet). Our method is dedicated to automatically selecting relevant soft proposals given weak annotations (e.g., global image labels), thereby significantly reducing the annotation burden, and encompasses the following contributions. First, the proposed WSCNet detects a sentiment-specific soft map by training a fully convolutional network with the cross spatial pooling strategy in the detection branch. Second, both the holistic and localized information are utilized by coupling the sentiment map with deep features as semantic vector in the classification branch. The sentiment detection and classification branches are integrated into a unified deep framework optimized in an end-to-end manner. Extensive experiments demonstrate that the proposed WSCNet outperforms the state-of-the-art results on seven benchmark datasets.

We present a two-stage approach to first constructing 3D panoramas and then stitching them for noise-resilient reconstruction of large-scale indoor scenes. Our approach requires multiple unsynchronized RGB-D cameras, mounted on a robot platform, which can perform in-place rotations at different locations in a scene. Such cameras rotate on a common (but unknown) axis, which provides a novel perspective for coping with unsynchronized cameras, without requiring sufficient overlap of their Field-of-View (FoV). Based on this key observation, we propose novel algorithms to track these cameras simultaneously. Furthermore, during the integration of raw frames onto an equirectangular panorama, we derive uncertainty estimates from multiple measurements assigned to the same pixels. This enables us to appropriately model the sensing noise and consider its influence, so as to achieve better noise resilience, and improve the geometric quality of each panorama and the accuracy of global inter-panorama registration. We evaluate and demonstrate the performance of our proposed method for enhancing the geometric quality of scene reconstruction from both real-world and synthetic scans.

We present a novel algorithm for semantic segmentation and labeling of 3D point clouds of indoor scenes, where objects in point clouds can have significant variations and complex configurations. Effective segmentation methods decomposing point clouds into semantically meaningful pieces are highly desirable for object recognition, scene understanding, scene modeling, etc. However, existing segmentation methods based on low-level geometry tend to either under-segment or over-segment point clouds. Our method takes a fundamentally different approach, where semantic segmentation is achieved along with labeling. To cope with substantial shape variation for objects in the same category, we first segment point clouds into surface patches and use unsupervised clustering to group patches in the training set into clusters, providing an intermediate representation for effectively learning patch relationships. During testing, we propose a novel patch segmentation and classification framework with multiscale processing, where the local segmentation level is automatically determined by exploiting the learned cluster based contextual information. Our method thus produces robust patch segmentation and semantic labeling results, avoiding parameter sensitivity. We further learn object-cluster relationships from the training set, and produce semantically meaningful object level segmentation. Our method outperforms state-of-the-art methods on several representative point cloud datasets, including S3DIS, SceneNN, Cornell RGB-D and ETH.

We present a novel approach to integrate data from multiple sensor types for dense 3D reconstruction of indoor scenes in realtime. Existing algorithms are mainly based on a single RGBD camera and thus require continuous scanning of areas with sufficient geometric features. Otherwise, tracking may fail due to unreliable frame registration. Inspired by the fact that the fusion of multiple sensors can combine their strengths towards a more robust and accurate self-localization, we incorporate multiple types of sensors which are prevalent in modern robot systems, including a 2D range sensor, an inertial measurement unit (IMU), and wheel encoders. We fuse their measurements to reinforce the tracking process and to eventually obtain better 3D reconstructions. Specifically, we develop a 2D truncated signed distance field (TSDF) volume representation for the integration and ray-casting of laser frames, leading to a unified cost function in the pose estimation stage. For validation of the estimated poses in the loop-closure optimization process, we train a classifier for the features extracted from heterogeneous sensors during the registration progress. To evaluate our method on challenging use case scenarios, we assembled a scanning platform prototype to acquire real-world scans. We further simulated synthetic scans based on high-fidelity synthetic scenes for quantitative evaluation. Extensive experimental evaluation on these two types of scans demonstrate that our system is capable of robustly acquiring dense 3D reconstructions and outperforms state-of-the-art RGBD and LiDAR systems.

Hand-drawn sketch recognition is a fundamental problem in computer vision, widely used in sketch-based image and video retrieval, editing, and reorganization. Previous methods often assume that a complete sketch is used as input; however, hand-drawn sketches in common application scenarios are often incomplete, which makes sketch recognition a challenging problem. In this paper, we propose SketchGAN, a new generative adversarial network (GAN) based approach that jointly completes and recognizes a sketch, boosting the performance of both tasks. Specifically, we use a cascade Encode-Decoder network to complete the input sketch in an iterative manner, and employ an auxiliary sketch recognition task to recognize the completed sketch. Experiments on the Sketchy database benchmark demonstrate that our joint learning approach achieves competitive sketch completion and recognition performance compared with the state-of-the-art methods. Further experiments using several sketch-based applications also validate the performance of our method.

Insect pests are one of the main factors affecting agricultural product yield. Accurate recognition of insect pests facilitates timely preventive measures to avoid economic losses. However, the existing datasets for the visual classification task mainly focus on common objects, e.g., flowers and dogs. This limits the application of powerful deep learning technology on specific domains like the agricultural field. In this paper, we collect a large-scale dataset named IP102 for insect pest recognition. Specifically, it contains more than 75, 000 images belonging to 102 categories, which exhibit a natural long-tailed distribution. In addition, we annotate about 19, 000 images with bounding boxes for object detection. The IP102 has a hierarchical taxonomy and the insect pests which mainly affect one specific agricultural product are grouped into the same upperlevel category. Furthermore, we perform several baseline experiments on the IP102 dataset, including handcrafted and deep feature based classification methods. Experimental results show that this dataset has the challenges of interand intra- class variance and data imbalance. We believe our IP102 will facilitate future research on practical insect pest control, fine-grained visual classification, and imbalanced learning fields. We make the dataset and pre-trained models publicly available at https://github.com/xpwu95/IP102.

Significant progress has been made with image stylization using deep learning, especially with generative adversarial networks (GANs). However, existing methods fail to produce high quality artistic portrait drawings. Such drawings have a highly abstract style, containing a sparse set of continuous graphical elements such as lines, and so small artifacts are much more exposed than for painting styles. Moreover, artists tend to use different strategies to draw different facial features and the lines drawn are only loosely related to obvious image features. To address these challenges, we propose APDrawingGAN, a novel GAN based architecture that builds upon hierarchical generators and discriminators combining both a global network (for images as a whole) and local networks (for individual facial regions). This allows dedicated drawing strategies to be learned for different facial features. Since artists' drawings may not have lines perfectly aligned with image features, we develop a novel loss to measure similarity between generated and artists' drawings based on distance transforms, leading to improved strokes in portrait drawing. To train APDrawingGAN, we construct an artistic drawing dataset containing high-resolution portrait photos and corresponding professional artistic drawings. Extensive experiments, including a user study, show that APDrawingGAN produces significantly better artistic drawings than state-of-the-art methods.

We present a practical backend for stereo visual SLAM which can simultaneously discover individual rigid bodies and compute their motions in dynamic environments. While recent factor graph based state optimization algorithms have shown their ability to robustly solve SLAM problems by treating dynamic objects as outliers, the dynamic motions are rarely considered. In this paper, we exploit the consensus of 3D motions among the landmarks extracted from the same rigid body for clustering and estimating static and dynamic objects in a unified manner. Specifically, our algorithm builds a noise-aware motion affinity matrix upon landmarks, and uses agglomerative clustering for distinguishing those rigid bodies. Accompanied by a decoupled factor graph optimization for revising their shape and trajectory, we obtain an iterative scheme to update both cluster assignments and motion estimation reciprocally. Evaluations on both synthetic scenes and KITTI demonstrate the capability of our approach, and further experiments considering online efficiency also show the effectiveness of our method for simultaneous tracking of ego-motion and multiple objects.

We present a novel algorithm for point cloud segmentation. Our approach transforms unstructured point clouds into regular voxel grids, and further uses a kernel-based interpolated variational autoencoder (VAE) architecture to encode the local geometry within each voxel. Traditionally, the voxel representation only comprises Boolean occupancy information which fails to capture the sparsely distributed points within voxels in a compact manner. In order to handle sparse distributions of points, we further employ radial basis functions (RBF) to compute a local, continuous representation within each voxel. Our approach results in a good volumetric representation that effectively tackles noisy point cloud datasets and is more robust for learning. Moreover, we further introduce group equivariant CNN to 3D, by defining the convolution operator on a symmetry group acting on Z3 and its isomorphic sets. This improves the expressive capacity without increasing parameters, leading to more robust segmentation results. We highlight the performance on standard benchmarks and show that our approach outperforms state-of-the-art segmentation algorithms on the ShapeNet and S3DIS datasets.

Accurate grading of skin disease severity plays a crucial role in precise treatment for patients. Acne vulgaris, the most common skin disease in adolescence, can be graded by evidence-based lesion counting as well as experience-based global estimation in the medical field. However, due to the appearance similarity of acne with close severity, it is challenging to count and grade acne accurately. In this paper, we address the problem of acne image analysis via Label Distribution Learning (LDL) considering the ambiguous information among acne severity. Based on the professional grading criterion, we generate two acne label distributions considering the relationship between the similar number of lesions and severity of acne, respectively. We also propose a unified framework for joint acne image grading and counting, which is optimized by the multi-task learning loss. In addition, we further build the ACNE04 dataset with annotations of acne severity and lesion number of each image for evaluation. Experiments demonstrate that our proposed framework performs favorably against state-of-the-art methods. We make the code and dataset publicly available at https://github.com/xpwu95/ldl.

Images play a crucial role for people to express their opinions online due to the increasing popularity of social networks. While an affective image retrieval system is useful for obtaining visual contents with desired emotions from a massive repository, the abstract and subjective characteristics make the task challenging. To address the problem, this paper introduces an Attention-aware Polarity Sensitive Embedding (APSE) network to learn affective representations in an end-to-end manner. First, to automatically discover and model the informative regions of interest, we develop a hierarchical attention mechanism, in which both polarity- and emotion-specific attended representations are aggregated for discriminative feature embedding. Second, we present a weighted emotion-pair loss to take the inter- and intra-polarity relationships of the emotional labels into consideration. Guided by attention module, we weight the sample pairs adaptively which further improves the performance of feature embedding. Extensive experiments on four popular benchmark datasets show that the proposed method performs favorably against the state-of-the-art approaches.

In this paper, we present a novel framework for global non-rigid registration of multi-view scans captured using consumer-level depth cameras. In our method, all scans from different viewpoints are allowed to undergo large non-rigid deformations and finally fused into a complete high quality model. To avoid the well-known loop closure problem, we simultaneously optimize a global alignment problem instead of pairwise non-rigid registration in succession. We employ a joint point-to-point and point-to-plane positional constraint to reduce the influence of wrong correspondences, and incorporate an as-conformal-as-possible constraint to avoid mesh distortions during deformation. We also design a reweighting scheme on position and transformation to reduce registration errors. Experimental results on both public datasets and real scanned datasets demonstrate that our approach outperforms state-of-the-art methods through extensive quantitative and qualitative evaluations.

Effective representation and reconstruction for human faces are very important in many applications. Existing linear representation methods cannot reconstruct high quality 3D faces with details, while the newest non-linear representation method is less suitable for real shapes since spectral decompositions are unstable across different graphs. To address these problems, we propose a multi-scale graph convolutional autoencoder for face representation and reconstruction. Our autoencoder uses graph convolution, which is easily trained for the data with graph structures and can be used for other deformable models. Our model can also be used for variational training to generate high quality face shapes. Experimental results demonstrate that our model can generate more plausible, complex, and stable 3D shapes, and achieves higher quality face reconstruction compared with state-of-the-art methods.

We present a real-time dense mapping system which uses the predicted 2D semantic labels for optimizing the geometric quality of reconstruction. With a combination of Convolutional Neural Networks (CNNs) for 2D labeling and a Simultaneous Localization and Mapping (SLAM) system for camera trajectory estimation, recent approaches have succeeded in incrementally fusing and labeling 3D scenes. However, the geometric quality of the reconstruction can be further improved by incorporating such semantic prediction results, which is not sufficiently exploited by existing methods. In this paper, we propose to use semantic information to improve two crucial modules in the reconstruction pipeline, namely tracking and loop detection, for obtaining mutual benefits in geometric reconstruction and semantic recognition. Specifically for tracking, we use a novel probabilistic projective association approach to efficiently pick out candidate correspondences, where the confidence of these correspondences is quantified concerning similarities on all available short-term invariant features. For the loop detection, we incorporate these semantic labels into the original encoding through Randomized Ferns to generate a more comprehensive representation for retrieving candidate loop frames. Evaluations on a publicly available synthetic dataset have shown the effectiveness of our approach that considers such semantic hints as a reliable feature for achieving higher geometric quality.

Non-rigid registration of deformed 3D shapes is a challenging and fundamental task in geometric processing, which aims to non-rigidly deform a source shape into alignment with a target shape. Current state-of-the-art methods assume deformations to be near-isometric. This assumption does not reflect real-world conditions, for example in large-scale deformation, where moderate anisotropic deformations (e.g., stretches) are common. In this paper we propose two significant changes to a typical registration pipeline to address such challenging deformations. First, we introduce a method to estimate anisotropic non-isometric deformations and incorporate this into an iterative non-rigid registration pipeline. Second, we compute additional correspondences in non-isometrically deforming regions using reliable correspondences as landmarks and prune inconsistent correspondences. We compare the performance of our proposed algorithm to several state-of-the-art methods using existing benchmarks. Experimental results show that our method outperforms existing methods.

Abstract In this work, we introduce multi‐column graph convolutional networks (MGCNs), a deep generative model for 3D mesh surfaces that effectively learns a non‐linear facial representation. We perform spectral decomposition of meshes and apply convolutions directly in the frequency domain. Our network architecture involves multiple columns of graph convolutional networks (GCNs), namely large GCN (L‐GCN), medium GCN (M‐GCN) and small GCN (S‐GCN), with different filter sizes to extract features at different scales. L‐GCN is more useful to extract large‐scale features, whereas S‐GCN is effective for extracting subtle and fine‐grained features, and M‐GCN captures information in between. Therefore, to obtain a high‐quality representation, we propose a selective fusion method that adaptively integrates these three kinds of information. Spatially non‐local relationships are also exploited through a self‐attention mechanism to further improve the representation ability in the latent vector space. Through extensive experiments, we demonstrate the superiority of our end‐to‐end framework in improving the accuracy of 3D face reconstruction. Moreover, with the help of variational inference, our model has excellent generating ability.

Given a reference colour image and a destination grayscale image, this paper presents a novel automatic colourisation algorithm that transfers colour information from the reference image to the destination image. Since the reference and destination images may contain content at different or even varying scales (due to changes of distance between objects and the camera), existing texture matching based methods can often perform poorly. We propose a novel cross-scale texture matching method to improve the robustness and quality of the colourisation results. Suitable matching scales are considered locally, which are then fused using global optimisation that minimises both the matching errors and spatial change of scales. The minimisation is efficiently solved using a multi-label graph-cut algorithm. Since only low-level texture features are used, texture matching based colourisation can still produce semantically incorrect results, such as meadow appearing above the sky. We consider a class of semantic violation where the statistics of up-down relationships learnt from the reference image are violated and propose an effective method to identify and correct unreasonable colourisation. Finally, a novel nonlocal ℓ1 optimisation framework is developed to propagate high confidence micro-scribbles to regions of lower confidence to produce a fully colourised image. Qualitative and quantitative evaluations show that our method outperforms several state-of-the-art methods.

We present a novel global non-rigid registration method for dynamic 3D objects. Our method allows objects to undergo large non-rigid deformations and achieves high-quality results even with substantial pose change or camera motion between views. In addition, our method does not require a template prior and uses less raw data than tracking-based methods since only a sparse set of scans is needed. We simultaneously compute the deformations of all the scans by optimizing a global alignment problem to avoid the well-known loop closure problem and use an as-rigid-as-possible constraint to eliminate the shrinkage problem of the deformed shapes, especially near open boundaries of scans. To cope with large-scale problems, we design a coarse-to-fine multi-resolution scheme, which also avoids the optimization being trapped into local minima. The proposed method is evaluated on public datasets and real datasets captured by an RGB-D sensor. The experimental results demonstrate that the proposed method obtains better results than several state-of-the-art methods.

We introduce SDM-NET, a deep generative neural network which produces structured deformable meshes. Specifically, the network is trained to generate a spatial arrangement of closed, deformable mesh parts, which respects the global part structure of a shape collection, e.g., chairs, airplanes, etc. Our key observation is that while the overall structure of a 3D shape can be complex, the shape can usually be decomposed into a set of parts, each homeomorphic to a box, and the finer-scale geometry of the part can be recovered by deforming the box. The architecture of SDM-NET is that of a two-level variational autoencoder (VAE). At the part level, a PartVAE learns a deformable model of part geometries. At the structural level, we train a Structured Parts VAE (SP-VAE), which jointly learns the part structure of a shape collection and the part geometries, ensuring the coherence between global shape structure and surface details. Through extensive experiments and comparisons with the state-of-the-art deep generative models of shapes, we demonstrate the superiority of SDM-NET in generating meshes with visual quality, flexible topology, and meaningful structures, benefiting shape interpolation and other subsequent modeling tasks.

Non-rigid registration is challenging because it is ill-posed with high degrees of freedom and is thus sensitive to noise and outliers. We propose a robust non-rigid registration method using reweighted sparsities on position and transformation to estimate the deformations between 3-D shapes. We formulate the energy function with position and transformation sparsity on both the data term and the smoothness term, and define the smoothness constraint using local rigidity. The double sparsity based non-rigid registration model is enhanced with a reweighting scheme, and solved by transferring the model into four alternately-optimized subproblems which have exact solutions and guaranteed convergence. Experimental results on both public datasets and real scanned datasets show that our method outperforms the state-of-the-art methods and is more robust to noise and outliers than conventional non-rigid registration methods.

Spatially localized deformation components are very useful for shape analysis and synthesis in 3D geometry processing. Several methods have recently been developed, with an aim to extract intuitive and interpretable deformation components. However, these techniques suffer from fundamental limitations especially for meshes with noise or large-scale deformations, and may not always be able to identify important deformation components.In this paper we propose a novel mesh-based autoencoder architecture that is able to cope with meshes with irregular topology. We introduce sparse regularization in this framework, which along with convolutional operations, helps localize deformations.Our framework is capable of extracting localized deformation components from mesh data sets with large-scale deformations and is robust to noise. It also provides a nonlinear approach to reconstruction of meshes using the extracted basis, which is more effective than the current linear combination approach. Extensive experiments show that our method outperforms state-of-the-art methods in both qualitative and quantitative evaluations.

Affective image understanding has been extensively studied in the last decade since more and more users express emotion via visual contents. While current algorithms based on convolutional neural networks aim to distinguish emotional categories in a discrete label space, the task is inherently ambiguous. This is mainly because emotional labels with the same polarity (i.e., positive or negative) are highly related, which is different from concrete object concepts such as cat, dog and bird. To the best of our knowledge, few methods focus on leveraging such characteristic of emotions for affective image understanding. In this work, we address the problem of understanding affective images via deep metric learning and propose a multi-task deep framework to optimize both retrieval and classification goals. We propose the sentiment constraints adapted from the triplet constraints, which are able to explore the hierarchical relation of emotion labels. We further exploit the sentiment vector as an effective representation to distinguish affective images utilizing the texture representation derived from convolutional layers. Extensive evaluations on four widely-used affective datasets, i.e., Flickr and Instagram, IAPSa, Art Photo, and Abstract Paintings, demonstrate that the proposed algorithm performs favorably against the state-of-the-art methods on both affective image retrieval and classification tasks.

In this paper, we propose a solution to transforming photos of real-world scenes into cartoon style images, which is valuable and challenging in computer vision and computer graphics. Our solution belongs to learning based methods, which have recently become popular to stylize images in artistic forms such as painting. However, existing methods do not produce satisfactory results for cartoonization, due to the fact that (1) cartoon styles have unique characteristics with high level simplification and abstraction, and (2) cartoon images tend to have clear edges, smooth color shading and relatively simple textures, which exhibit significant challenges for texture-descriptor-based loss functions used in existing methods. In this paper, we propose CartoonGAN, a generative adversarial network (GAN) framework for cartoon stylization. Our method takes unpaired photos and cartoon images for training, which is easy to use. Two novel losses suitable for cartoonization are proposed: (1) a semantic content loss, which is formulated as a sparse regularization in the high-level feature maps of the VGG network to cope with substantial style variation between photos and cartoons, and (2) an edge-promoting adversarial loss for preserving clear edges. We further introduce an initialization phase, to improve the convergence of the network to the target manifold. Our method is also much more efficient to train than existing methods. Experimental results show that our method is able to generate high-quality cartoon images from real-world photos (i.e., following specific artists' styles and with clear edges and smooth shading) and outperforms state-of-the-art methods.

3D geometric contents are becoming increasingly popular. In this paper, we study the problem of analyzing deforming 3D meshes using deep neural networks. Deforming 3D meshes are flexible to represent 3D animation sequences as well as collections of objects of the same category, allowing diverse shapes with large-scale non-linear deformations. We propose a novel framework which we call mesh variational autoencoders (mesh VAE), to explore the probabilistic latent space of 3D surfaces. The framework is easy to train, and requires very few training examples. We also propose an extended model which allows flexibly adjusting the significance of different latent variables by altering the prior distribution. Extensive experiments demonstrate that our general framework is able to learn a reasonable representation for a collection of deformable shapes, and produce competitive results for a variety of applications, including shape generation, shape interpolation, shape space embedding and shape exploration, outperforming state-of-the-art methods.

Caricature is an art form that expresses subjects in abstract, simple and exaggerated views. While many caricatures are 2D images, this paper presents an algorithm for creating expressive 3D caricatures from 2D caricature images with minimum user interaction. The key idea of our approach is to introduce an intrinsic deformation representation that has the capability of extrapolation, enabling us to create a deformation space from standard face datasets, which maintains face constraints and meanwhile is sufficiently large for producing exaggerated face models. Built upon the proposed deformation representation, an optimization model is formulated to find the 3D caricature that captures the style of the 2D caricature image automatically. The experiments show that our approach has better capability in expressing caricatures than those fitting approaches directly using classical parametric face models such as 3DMM and FaceWareHouse. Moreover, our approach is based on standard face datasets and avoids constructing complicated 3D caricature training sets, which provides great flexibility in real applications.

Automatic assessment of sentiment from visual content has gained considerable attention with the increasing tendency of expressing opinions on-line. In this paper, we solve the problem of visual sentiment analysis using the high-level abstraction in the recognition process. Existing methods based on convolutional neural networks learn sentiment representations from the holistic image appearance. However, different image regions can have a different influence on the intended expression. This paper presents a weakly supervised coupled convolutional network with two branches to leverage the localized information. The first branch detects a sentiment specific soft map by training a fully convolutional network with the cross spatial pooling strategy, which only requires image-level labels, thereby significantly reducing the annotation burden. The second branch utilizes both the holistic and localized information by coupling the sentiment map with deep features for robust classification. We integrate the sentiment detection and classification branches into a unified deep framework and optimize the network in an end-to-end manner. Extensive experiments on six benchmark datasets demonstrate that the proposed method performs favorably against the state-of-the-art methods for visual sentiment analysis.

Supervoxels are perceptually meaningful atomic regions in videos, obtained by grouping voxels that exhibit coherence in both appearance and motion. In this paper, we propose content-sensitive supervoxels (CSS), which are regularly-shaped 3D primitive volumes that possess the following characteristic: they are typically larger and longer in content-sparse regions (i.e., with homogeneous appearance and motion), and smaller and shorter in content-dense regions (i.e., with high variation of appearance and/or motion). To compute CSS, we map a video Ξ to a 3-dimensional manifold M embedded in R6, whose volume elements give a good measure of the content density in Ξ. We propose an efficient Lloyd-like method with a splitting-merging scheme to compute a uniform tessellation on M, which induces the CSS in Ξ. Theoretically our method has a good competitive ratio O(1). We also present a simple extension of CSS to stream CSS for processing long videos that cannot be loaded into main memory at once. We evaluate CSS, stream CSS and seven representative supervoxel methods on four video datasets. The results show that our method outperforms existing supervoxel methods.

Micro-expression (ME) is a special type of human expression which can reveal the real emotion that people want to conceal. Spontaneous ME (SME) spotting is to identify the subsequences containing SMEs from a long facial video. The study of SME spotting has a significant importance, but is also very challenging due to the fact that in real-world scenarios, SMEs may occur along with normal facial expressions and other prominent motions such as head movements. In this paper, we improve a state-of-the-art SME spotting method called feature difference analysis (FD) in the following two aspects. First, FD relies on a partitioning of facial area into uniform regions of interest (ROIs) and computing features of a selected sequence. We propose a novel evaluation method by utilizing the Fisher linear discriminant to assign a weight for each ROI, leading to more semantically meaningful ROIs. Second, FD only considers two features (LBP and HOOF) independently. We introduce a state-of-the-art MDMO feature into FD and propose a simple yet efficient collaborative strategy to work with two complementary features, i.e., LBP characterizing texture information and MDMO characterizing motion information. We call our improved FD method collaborative feature difference (CFD). Experimental results on two well-established SME datasets SMIC-E and CASME II show that CFD significantly improves the performance of the original FD.

Supervoxels are perceptually meaningful atomic spatiotemporal regions in videos, which has great potential to reduce the computational complexity of downstream video applications. Many methods have been proposed for generating supervoxels. To effectively evaluate these methods, a novel supervoxel library and benchmark called LIBSVX with seven collected metrics was recently established. In this paper, we propose a new compactness metric which measures the shape regularity of supervoxels and is served as a necessary complement to the existing metrics. To demonstrate its necessity, we first explore the relations between the new metric and existing ones. Correlation analysis shows that the new metric has a weak correlation with (i.e., nearly independent of) existing metrics, and so reflects a new characteristic of supervoxel quality. Second, we investigate two real-world video applications. Experimental results show that the new metric can effectively predict some important application performance, while most existing metrics cannot do so.

With the rapid development of virtual reality (VR) technology, VR glasses, a.k.a. Head-Mounted Displays (HMDs) are widely available, allowing immersive 3D content to be viewed. A natural need for truly immersive VR is to allow bidirectional communication: the user should be able to interact with the virtual world using facial expressions and eye gaze, in addition to traditional means of interaction. Typical application scenarios include VR virtual conferencing and virtual roaming, where ideally users are able to see other users' expressions and have eye contact with them in the virtual world. Despite significant achievements in recent years for reconstruction of 3D faces from RGB or RGB- D images, it remains a challenge to reliably capture and reconstruct 3D facial expressions including eye gaze when the user is wearing VR glasses, because the majority of the face is occluded, especially those areas around the eyes which are essential for recognizing facial expressions and eye gaze. In this paper, we introduce a novel real-time system that is able to capture and reconstruct 3D faces wearing HMDs and robustly recover eye gaze. We demonstrate the effectiveness of our system using live capture and more results are shown in the accompanying video.

We consider the problem of learning a representation of both spatial relations and dependencies between objects for indoor scene design. We propose a novel knowledge graph framework based on the entity-relation model for representation of facts in indoor scene design, and further develop a weaklysupervised algorithm for extracting the knowledge graph representation from a small dataset using both structure and parameter learning. The proposed framework is flexible, transferable, and readable. We present a variety of computer-aided indoor scene design applications using this representation, to show the usefulness and robustness of the proposed framework.

Benefiting from global rank constraints, the low-rank representation (LRR) method has been shown to be an effective solution to subspace learning. However, the global mechanism also means that the LRR model is not suitable for handling large-scale data or dynamic data. For large-scale data, the LRR method suffers from high time complexity, and for dynamic data, it has to recompute a complex rank minimization for the entire data set whenever new samples are dynamically added, making it prohibitively expensive. Existing attempts to online LRR either take a stochastic approach or build the representation purely based on a small sample set and treat new input as out-of-sample data. The former often requires multiple runs for good performance and thus takes longer time to run, and the latter formulates online LRR as an out-of-sample classification problem and is less robust to noise. In this paper, a novel online LRR subspace learning method is proposed for both large-scale and dynamic data. The proposed algorithm is composed of two stages: static learning and dynamic updating. In the first stage, the subspace structure is learned from a small number of data samples. In the second stage, the intrinsic principal components of the entire data set are computed incrementally by utilizing the learned subspace structure, and the LRR matrix can also be incrementally solved by an efficient online singular value decomposition algorithm. The time complexity is reduced dramatically for large-scale data, and repeated computation is avoided for dynamic problems. We further perform theoretical analysis comparing the proposed online algorithm with the batch LRR method. Finally, experimental results on typical tasks of subspace recovery and subspace clustering show that the proposed algorithm performs comparably or better than batch methods, including the batch LRR, and significantly outperforms state-of-the-art online methods.

We develop a framework to virtually unroll fragile historical parchment scrolls, which cannot be physically unfolded via a sequence of X-ray tomographic slices, thus providing easy access to those parchments whose contents have remained hidden for centuries. The first step is to produce a topologically correct segmentation, which is challenging as the parchment layers vary significantly in thickness, contain substantial interior textures and can often stick together in places. For this purpose, our method starts with linking the broken layers in a slice using the topological structure propagated from its previous processed slice. To ensure topological correctness, we identify fused regions by detecting junction sections, and then match them using global optimization efficiently solved by the blossom algorithm, taking into account the shape energy of curves separating fused layers. The fused layers are then separated using as-parallel-as-possible curves connecting junction section pairs. To flatten the segmented parchment, pixels in different frames need to be put into alignment. This is achieved via a dynamic programming-based global optimization, which minimizes the total matching distances and penalizes stretches. Eventually, the text of the parchment is revealed by ink projection. We demonstrate the effectiveness of our approach using challenging real-world data sets, including the water damaged fifteenth century Bressingham scroll.

Leveraging the abundant number of web data is a promising strategy in addressing the problem of data lacking when training convolutional neural networks (CNNs). However, the web images often contain incorrect tags, which may compromise the learned CNN model. To address this problem, this paper focuses on image classification and proposes to iterate between filtering out noisy web labels and fine-tuning the CNN model using the crawled web images. Overall, the proposed method benefits from the growing modeling capability of the learned model to correct labels for web images and learning from such new data to produce a more effective model. Our contribution is two-fold. First, we propose an iterative method that progressively improves the discriminative ability of CNNs and the accuracy of web image selection. This method is beneficial toward selecting high-quality web training images and expanding the training set as the model gets ameliorated. Second, since web images are usually complex and may not be accurately described by a single tag, we propose to assign a web image multiple labels to reduce the impact of hard label assignment. This labeling strategy mines more training samples to improve the CNN model. In the experiments, we crawl 0.5 million web images covering all categories of four public image classification data sets. Compared with the baseline which has no web images for training, we show that the proposed method brings notable improvement. We also report the competitive recognition accuracy compared with the state of the art.

Transferring deformation from a source shape to a target shape is a very useful technique in computer graphics. State-of-the-art deformation transfer methods require either point-wise correspondences between source and target shapes, or pairs of deformed source and target shapes with corresponding deformations. However, in most cases, such correspondences are not available and cannot be reliably established using an automatic algorithm. Therefore, substantial user effort is needed to label the correspondences or to obtain and specify such shape sets. In this work, we propose a novel approach to automatic deformation transfer between two unpaired shape sets without correspondences. 3D deformation is represented in a high-dimensional space. To obtain a more compact and effective representation, two convolutional variational autoencoders are learned to encode source and target shapes to their latent spaces. We exploit a Generative Adversarial Network (GAN) to map deformed source shapes to deformed target shapes, both in the latent spaces, which ensures the obtained shapes from the mapping are indistinguishable from the target shapes. This is still an under-constrained problem, so we further utilize a reverse mapping from target shapes to source shapes and incorporate cycle consistency loss, i.e. applying both mappings should reverse to the input shape. This VAE-Cycle GAN (VC-GAN) architecture is used to build a reliable mapping between shape spaces. Finally, a similarity constraint is employed to ensure the mapping is consistent with visual similarity, achieved by learning a similarity neural network that takes the embedding vectors from the source and target latent spaces and predicts the light field distance between the corresponding shapes. Experimental results show that our fully automatic method is able to obtain high-quality deformation transfer results with unpaired data sets, comparable or better than existing methods where strict correspondences are required.

Image retargeting techniques that adjust images into different sizes have attracted much attention recently. Objective quality assessment (OQA) of image retargeting results is often desired to automatically select the best results. Existing OQA methods output an absolute score for each retargeted image and use these scores to compare different results. Observing that it is challenging even for human subjects to give consistent scores for retargeting results of different source images, in this paper we propose a learning-based OQA method that predicts the ranking of a set of retargeted images with the same source image. We show that this more manageable task helps achieve more consistent prediction to human preference and is sufficient for most application scenarios. To compute the ranking, we propose a simple yet efficient machine learning framework that uses a General Regression Neural Network (GRNN) to model a combination of seven elaborate OQA metrics. We then propose a simple scheme to transform the relative scores output from GRNN into a global ranking. We train our GRNN model using human preference data collected in the elaborate RetargetMe benchmark and evaluate our method based on the subjective study in RetargetMe. Moreover, we introduce a further subjective benchmark to evaluate the generalizability of different OQA methods. Experimental results demonstrate that our method outperforms eight representative OQA methods in ranking prediction and has better generalizability to different datasets.

In this paper, inspired by Gatys's recent work, we propose a novel approach that transforms photos to comics using deep convolutional neural networks (CNNs). While Gatys's method that uses a pre-trained VGG network generally works well for transferring artistic styles such as painting from a style image to a content image, for more minimalist styles such as comics, the method often fails to produce satisfactory results. To address this, we further introduce a dedicated comic style CNN, which is trained for classifying comic images and photos. This new network is effective in capturing various comic styles and thus helps to produce better comic stylization results. Even with a grayscale style image, Gatys's method can still produce colored output, which is not desirable for comics. We develop a modified optimization framework such that a grayscale image is guaranteed to be synthesized. To avoid converging to poor local minima, we further initialize the output image using grayscale version of the content image. Various examples show that our method synthesizes better comic images than the state-of-the-art method.

We present a novel global registration method for deformable objects captured using a single RGB-D camera. Our algorithm allows objects to undergo large non-rigid deformations, and achieves high quality results without constraining the actor's pose or camera motion. We compute the deformations of all the scans simultaneously by optimizing a global alignment problem to avoid the well-known loop closure problem, and use an as-rigid-as-possible constraint to eliminate the shrinkage problem of the deformed model. To attack large scale problems, we design a coarse-to-fine multi-resolution scheme, which also avoids the optimization being trapped into local minima. The proposed method is evaluated on public datasets and real datasets captured by an RGB-D sensor. Experimental results demonstrate that the proposed method obtains better results than the state-of-the-art methods.

This paper addresses the challenge of 3-D skeleton recovery by exploiting the spatio-temporal correlations of corrupted 3D skeleton sequences. A skeleton sequence is represented as a matrix. We propose a novel low-rank solution that effectively integrates both a low-rank model for robust skeleton recovery based on temporal coherence, and an articulation-graph-based isometric constraint for spatial coherence, namely consistency of bone lengths. The proposed model is formulated as a constrained optimization problem, which is efficiently solved by the Augmented Lagrangian Method with a Gauss-Newton solver for the subproblem of isometric optimization. Experimental results on the CMU motion capture dataset and a Kinect dataset show that the proposed approach achieves better recovery accuracy over a state-of-the-art method. The proposed method has wide applicability for skeleton tracking devices, such as the Kinect, because these devices cannot provide accurate reconstructions of complex motions, especially in the presence of occlusion.

Abstract Shape interpolation has many applications in computer graphics such as morphing for computer animation. In this paper, we propose a novel data‐driven mesh interpolation method. We adapt patch‐based linear rotational invariant coordinates to effectively represent deformations of models in a shape collection, and utilize this information to guide the synthesis of interpolated shapes. Unlike previous data‐driven approaches, we use a rotation/translation invariant representation which defines the plausible deformations in a global continuous space. By effectively exploiting the knowledge in the shape space, our method produces realistic interpolation results at interactive rates, outperforming state‐of‐the‐art methods for challenging cases. We further propose a novel approach to interactive editing of shape morphing according to the shape distribution. The user can explore the morphing path and select example models intuitively and adjust the path with simple interactions to edit the morphing sequences. This provides a useful tool to allow users to generate desired morphing with little effort. We demonstrate the effectiveness of our approach using various examples.

Photorealistic animation is a desirable technique for computer games and movie production. We propose a new method to synthesize plausible videos of human actors with new motions using a single cheap RGB-D camera. A small database is captured in a usual office environment, which happens only once for synthesizing different motions. We propose a marker-less performance capture method using sparse deformation to obtain the geometry and pose of the actor for each time instance in the database. Then, we synthesize an animation video of the actor performing the new motion that is defined by the user. An adaptive model-guided texture synthesis method based on weighted low-rank matrix completion is proposed to be less sensitive to noise and outliers, which enables us to easily create photorealistic animation videos with new motions that are different from the motions in the database. Experimental results on the public data set and our captured data set have verified the effectiveness of the proposed method.

Image colorization aims to produce a natural looking color image from a given gray-scale image, which remains a challenging problem. In this paper, we propose a novel example-based image colorization method exploiting a new locality consistent sparse representation. Given a single reference color image, our method automatically colorizes the target gray-scale image by sparse pursuit. For efficiency and robustness, our method operates at the superpixel level. We extract low-level intensity features, mid-level texture features, and high-level semantic features for each superpixel, which are then concatenated to form its descriptor. The collection of feature vectors for all the superpixels from the reference image composes the dictionary. We formulate colorization of target superpixels as a dictionary-based sparse reconstruction problem. Inspired by the observation that superpixels with similar spatial location and/or feature representation are likely to match spatially close regions from the reference image, we further introduce a locality promoting regularization term into the energy formulation, which substantially improves the matching consistency and subsequent colorization results. Target superpixels are colorized based on the chrominance information from the dominant reference superpixels. Finally, to further improve coherence while preserving sharpness, we develop a new edge-preserving filter for chrominance channels with the guidance from the target gray-scale image. To the best of our knowledge, this is the first work on sparse pursuit image colorization from single reference images. Experimental results demonstrate that our colorization method outperforms the state-of-the-art methods, both visually and quantitatively using a user study.

Human action recognition from videos has wide applicability and receives significant interests. In this work, to better identify spatio-temporal characteristics, we propose a novel 3D extension of Gradient Location and Orientation Histograms, which provides discriminative local features representing not only the gradient orientation, but also their relative locations. We further propose a human action recognition system based on the Bag of Visual Words model, by combining the new 3D GLOH local features with Histograms of Oriented Optical Flow (HOOF) global features. Along with the idea from our recent work to extract features only in salient regions, our overall system outperforms existing feature descriptors for human action recognition for challenging real-world video datasets.

Skeleton tracking is a useful and popular application of Kinect. However, it cannot provide accurate reconstructions for complex motions, especially in the presence of occlusion. This paper proposes a new 3-D motion recovery method based on low-rank matrix analysis to correct invalid or corrupted motions. We address this problem by representing a motion sequence as a matrix, and introducing a convex low-rank matrix recovery model, which fixes erroneous entries and finds the correct low-rank matrix by minimizing nuclear norm and norm of constituent clean motion and error matrices. Experimental results show that our method recovers the corrupted skeleton joints, achieving accurate and smooth reconstructions even for complicated motions.

Shape registration is fundamental to 3D object acquisition; it is used to fuse scans from multiple views. Existing algorithms mainly utilize geometric information to determine alignment, but this typically results in noticeable misalignment of textures (i.e. surface colors) when using RGB-depth cameras. We address this problem using a novel approach to color-aware registration, which takes both color and geometry into consideration simultaneously. Color information is exploited throughout the pipeline to provide more effective sampling, correspondence and alignment, in particular for surfaces with detailed textures. Our method can furthermore tackle both rigid and non-rigid registration problems (arising, for example, due to small changes in the object during scanning, or camera distortions). We demonstrate that our approach produces significantly better results than previous methods.

This paper presents a novel framework for human action recognition based on salient object detection and a new combination of local and global descriptors. We first detect salient objects in video frames and only extract features for such objects. We then use a simple strategy to identify and process only those video frames that contain salient objects. Processing salient objects instead of all frames not only makes the algorithm more efficient, but more importantly also suppresses the interference of background pixels. We combine this approach with a new combination of local and global descriptors, namely 3D-SIFT and histograms of oriented optical flow (HOOF), respectively. The resulting saliency guided 3D-SIFT–HOOF (SGSH) feature is used along with a multi-class support vector machine (SVM) classifier for human action recognition. Experiments conducted on the standard KTH and UCF-Sports action benchmarks show that our new method outperforms the competing state-of-the-art spatiotemporal feature-based human action recognition methods.

Effectively characterizing the behavior of deformable objects has wide applicability but remains challenging. We present a new rotation-invariant deformation representation and a novel reconstruction algorithm to accurately reconstruct the positions and local rotations simultaneously. Meshes can be very efficiently reconstructed from our representation by matrix pre-decomposition, while, at the same time, hard or soft constraints can be flexibly specified with only positions of handles needed. Our approach is thus particularly suitable for constrained deformations guided by examples, providing significant benefits over state-of-the-art methods. Based on this, we further propose novel data-driven approaches to mesh deformation and non-rigid registration of deformable objects. Both problems are formulated consistently as finding an optimized model in the shape space that satisfies boundary constraints, either specified by the user, or according to the scan. By effectively exploiting the knowledge in the shape space, our method produces realistic deformation results in real-time and produces high quality registrations from a template model to a single noisy scan captured using a low-quality depth camera, outperforming state-of-the-art methods.

Trademarks are signs of high reputational value. Thus, they require protection. This paper studies conceptual similarities between trademarks, which occurs when two or more trademarks evoke identical or analogous semantic content. This paper advances the state-of-the-art by proposing a computational approach based on semantics that can be used to compare trademarks for conceptual similarity. A trademark retrieval algorithm is developed that employs natural language processing techniques and an external knowledge source in the form of a lexical ontology. The search and indexing technique developed uses similarity distance, which is derived using Tversky's theory of similarity. The proposed retrieval algorithm is validated using two resources: a trademark database of 1400 disputed cases and a database of 378 943 company names. The accuracy of the algorithm is estimated using measures from two different domains: the R-precision score, which is commonly used in information retrieval and human judgment/collective human opinion, which is used in human-machine systems.

Abstract In vector graphics, gradient meshes represent an image object by one or more regularly connected grids. Every grid point has attributes as the position, colour and gradients of these quantities specified. Editing the attributes of an existing gradient mesh (such as the colour gradients) is not only non‐intuitive but also time‐consuming. To facilitate user‐friendly colour editing, we develop an optimization‐based colour transfer method for gradient meshes. The key idea is built on the fact that we can approximate a colour transfer operation on gradient meshes with a linear transfer function. In this paper, we formulate the approximation as an optimization problem, which aims to minimize the colour distribution of the example image and the transferred gradient mesh. By adding proper constraints, i.e. image gradients, to the optimization problem, the details of the gradient meshes can be better preserved. With the linear transfer function, we are able to edit the colours and colour gradients of the mesh points automatically, while preserving the structure of the gradient mesh. The experimental results show that our method can generate pleasing recoloured gradient meshes.

Abstract Non‐rigid registration of 3D shapes is an essential task of increasing importance as commodity depth sensors become more widely available for scanning dynamic scenes. Non‐rigid registration is much more challenging than rigid registration as it estimates a set of local transformations instead of a single global transformation, and hence is prone to the overfitting issue due to underdetermination. The common wisdom in previous methods is to impose an ℓ2‐norm regularization on the local transformation differences. However, the ℓ2‐norm regularization tends to bias the solution towards outliers and noise with heavy‐tailed distribution, which is verified by the poor goodness‐of‐fit of the Gaussian distribution over transformation differences. On the contrary, Laplacian distribution fits well with the transformation differences, suggesting the use of a sparsity prior. We propose a sparse non‐rigid registration (SNR) method with an ℓ1‐norm regularized model for transformation estimation, which is effectively solved by an alternate direction method (ADM) under the augmented Lagrangian framework. We also devise a multi‐resolution scheme for robust and progressive registration. Results on both public datasets and our scanned datasets show the superiority of our method, particularly in handling large‐scale deformations as well as outliers and noise.

3D scene modeling has long been a fundamental problem in computer graphics and computer vision. With the popularity of consumer-level RGB-D cameras, there is a growing interest in digitizing real-world indoor 3D scenes. However, modeling indoor 3D scenes remains a challenging problem because of the complex structure of interior objects and poor quality of RGB-D data acquired by consumer-level sensors. Various methods have been proposed to tackle these challenges. In this survey, we provide an overview of recent advances in indoor scene modeling techniques, as well as public datasets and code libraries which can facilitate experiments and evaluation.

With broader availability of large-scale 3D model repositories, the need for efficient and effective exploration becomes more and more urgent. Existing model retrieval techniques do not scale well with the size of the database since often a large number of very similar objects are returned for a query, and the possibilities to refine the search are quite limited. We propose an interactive approach where the user feeds an active learning procedure by labeling either entire models or parts of them as "like" or "dislike" such that the system can automatically update an active set of recommended models. To provide an intuitive user interface, candidate models are presented based on their estimated relevance for the current query. From the methodological point of view, our main contribution is to exploit not only the similarity between a query and the database models but also the similarities among the database models themselves. We achieve this by an offline pre-processing stage, where global and local shape descriptors are computed for each model and a sparse distance metric is derived that can be evaluated efficiently even for very large databases. We demonstrate the effectiveness of our method by interactively exploring a repository containing over 100 K models.

Trademarks are visual symbols with high reputational value, which requires protection. This paper proposes an algorithm to retrieve phonetically similar trademarks that can be used as a means for supporting trademark examination during the registration process. The algorithm employs a phonology based string similarity algorithm together with a typography mapping and token rearrangement to compute a phonetic similarity between trademarks. The trademark phonetic similarity score is then computed from the employed phonetic similarity algorithm. The proposed algorithm advances the state-of-the-art in trademark retrieval by providing a mechanism to compare trademarks with special characters or symbols phonetically. The proposed algorithm is tested on 1,400 trademarks obtained from real court cases between 1999 and 2012. The proposed algorithm improves the R-precision score by 14% and 17% compared with two state-of-the-art methods.

Abstract Identifying multiple deformable parts on meshes and establishing dense correspondences between them are tasks of fundamental importance to computer graphics, with applications to e.g. geometric edit propagation and texture transfer. Much research has considered establishing correspondences between non‐rigid surfaces, but little work can both identify similar multiple deformable partsandhandle partial shape correspondences. This paper addresses two related problems, treating them as a whole: (i) identifying similar deformable parts on a mesh, related by anon‐rigidtransformation to a given query part, and (ii) establishing dense point correspondences automatically between such parts. We show that simple and efficient techniques can be developed if we make the assumption that these parts locally undergo isometric deformation. Our insight is that similar deformable parts are suggested by large clusters of point correspondences that are isometrically consistent. Once such parts are identified,densepoint correspondences can be obtained by an iterative propagation process. Our techniques are applicable to models with arbitrary topology. Various examples demonstrate the effectiveness of our techniques.

Otsu's algorithm for thresholding images is widely used, and the computational complexity of determining the threshold from the histogram is O(N) where N is the number of histogram bins. When the algorithm is adapted to circular rather than linear histograms then two thresholds are required for binary thresholding. We show that, surprisingly, it is still possible to determine the optimal threshold in O(N) time. The efficient optimal algorithm is over 300 times faster than traditional approaches for typical histograms and is thus particularly suitable for real-time applications. We further demonstrate the usefulness of circular thresholding using the adapted Otsu criterion for various applications, including analysis of optical flow data, indoor/outdoor image classification, and non-photorealistic rendering. In particular, by combining circular Otsu feature with other colour/texture features, a 96.9% correct rate is obtained for indoor/outdoor classification on the well known IITM-SCID2 data set, outperforming the state-of-the-art result by 4.3%.

We present a novel solution to automatic semantic modeling of indoor scenes from a sparse set of low-quality RGB-D images. Such data presents challenges due to noise, low resolution, occlusion and missing depth information. We exploit the knowledge in a scene database containing 100s of indoor scenes with over 10,000 manually segmented and labeled mesh models of objects. In seconds, we output a visually plausible 3D scene, adapting these models and their parts to fit the input scans. Contextual relationships learned from the database are used to constrain reconstruction, ensuring semantic compatibility between both object models and parts. Small objects and objects with incomplete depth information which are difficult to recover reliably are processed with a two-stage approach. Major objects are recognized first, providing a known scene structure. 2D contour-based model retrieval is then used to recover smaller objects. Evaluations using our own data and two public datasets show that our approach can model typical real-world indoor scenes efficiently and robustly.

Finding correspondences between two surfaces is a fundamental operation in various applications in computer graphics and related fields. Candidate correspondences can be found by matching local signatures, but as they only consider local geometry, many are globally inconsistent. We provide a novel algorithm to prune a set of candidate correspondences to those most likely to be globally consistent. Our approach can handle articulated surfaces, and ones related by a deformation which is globally nonisometric, provided that the deformation is locally approximately isometric. Our approach uses an efficient diffusion framework, and only requires geodesic distance calculations in small neighbourhoods, unlike many existing techniques which require computation of global geodesic distances. We demonstrate that, for typical examples, our approach provides significant improvements in accuracy, yet also reduces time and memory costs by a factor of several hundred compared to existing pruning techniques. Our method is furthermore insensitive to holes, unlike many other methods.

Filling a small hole in an image with plausible content is well studied. Extrapolating an image to give a distinctly larger one is much more challenging---a significant amount of additional content is needed which matches the original image, especially near its boundaries. We propose a data-driven approach to this problem. Given a source image, and the amount and direction(s) in which it is to be extrapolated, our system determines visually consistent content for the extrapolated regions using library images. As well as considering low-level matching, we achieve consistency at a higher level by using graph proxies for regions of source and library images. Treating images as graphs allows us to find candidates for image extrapolation in a feasible time. Consistency of subgraphs in source and library images is used to find good candidates for the additional content; these are then further filtered. Region boundary curves are aligned to ensure consistency where image parts are joined using a photomontage method. We demonstrate the power of our method in image editing applications.

In this paper, we propose a data-driven approach for realistic deformation of articulated meshes: the wide availability of pose data makes such an approach feasible. We use multiple reference models, combined with a fast linear blending skinning (LBS) method in local spaces. By exploiting information from the reference models, realistic deformation is achieved, yet with computational efficiency comparable to traditional blending methods-our approach is suitable for real-time applications. We demonstrate the effectiveness of our algorithm using various examples.

The rapid expansion of e-commerce at the beginning of 21st century has had a significant impact on intellectual property management. A particular area of concern is the misuse of trademarks and trademark protection. Trademarks are proprietary words and images with high reputational value; they are important assets, often used as a marketing tool, which require infringement protection. One of the issues considered during infringement litigation is the visual, conceptual and phonetic similarity of different trademarks. In particular, the conceptual similarity of trademarks is an area never previously studied in information retrieval. This paper focuses on this important aspect by proposing a conceptual model of the comparison process, aimed at retrieving conceptually similar trademarks. The proposed model employs natural language processing and semantic technology to compute the conceptual similarity between trademarks.

Abstract Morphing between 3D objects is a fundamental technique in computer graphics. Traditional methods of shape morphing focus on establishing meaningful correspondences and finding smooth interpolation between shapes. Such methods however only take geometric information as input and thus cannot in general avoid producing unnatural interpolation, in particular for large‐scale deformations. This paper proposes a novel data‐driven approach for shape morphing. Given a database with various models belonging to the same category, we treat them as data samples in the plausible deformation space. These models are then clustered to form local shape spaces of plausible deformations. We use a simple metric to reasonably represent the closeness between pairs of models. Given source and target models, the morphing problem is casted as a global optimization problem of finding a minimal distance path within the local shape spaces connecting these models. Under the guidance of intermediate models in the path, an extended as‐rigid‐as‐possible interpolation is used to produce the final morphing. By exploiting the knowledge of plausible models, our approach produces realistic morphing for challenging cases as demonstrated by various examples in the paper.

Editing a photo-realistic gradient mesh is a tough task. Even only editing the colors of an existing gradient mesh can be exhaustive and time-consuming. To facilitate user-friendly color editing, we develop an example-based color transfer method for gradient meshes, which borrows the color characteristics of an example image to a gradient mesh. We start by exploiting the constraints of the gradient mesh, and accordingly propose a linear-operator-based color transfer framework. Our framework operates only on colors and color gradients of the mesh points and preserves the topological structure of the gradient mesh. Bearing the framework in mind, we build our approach on PCA-based color transfer. After relieving the color range problem, we incorporate a fusion-based optimization scheme to improve color similarity between the reference image and the recolored gradient mesh. Finally, a multi-swatch transfer scheme is provided to enable more user control. Our approach is simple, effective, and much faster than color transferring the rastered gradient mesh directly. The experimental results also show that our method can generate pleasing recolored gradient meshes.

Feature matching is a challenging problem at the heart of numerous computer graphics and computer vision applications. We present the SuperMatching algorithm for finding correspondences between two sets of features. It does so by considering triples or higher order tuples of points, going beyond the pointwise and pairwise approaches typically used. SuperMatching is formulated using a supersymmetric tensor representing an affinity metric that takes into account feature similarity and geometric constraints between features: Feature matching is cast as a higher order graph matching problem. SuperMatching takes advantage of supersymmetry to devise an efficient sampling strategy to estimate the affinity tensor, as well as to store the estimated tensor compactly. Matching is performed by an efficient higher order power iteration approach that takes advantage of this compact representation. Experiments on both synthetic and real data show that SuperMatching provides more accurate feature matching than other state-of-the-art approaches for a wide range of 2D and 3D features, with competitive computational cost.

We introduce a novel stratified sampling technique for mesh surfaces that gives the user control over sampling density and anisotropy via a tensor field. Our approach is based on sampling space-filling curves mapped onto mesh segments via parametrizations aligned with the tensor field. After a short preprocessing step, samples can be generated in real time. Along with visual examples, we provide rigorous spectral analysis and differential domain analysis of our sampling. The sample distributions are of high quality: they fulfil the blue noise criterion, so have minimal artifacts due to regularity of sampling patterns, and they accurately represent isotropic and anisotropic densities on the plane and on mesh surfaces. They also have low discrepancy, ensuring that the surface is evenly covered.

Three-dimensional surface registration transforms multiple three-dimensional data sets into the same coordinate system so as to align overlapping components of these sets. Recent surveys have covered different aspects of either rigid or nonrigid registration, but seldom discuss them as a whole. Our study serves two purposes: 1) To give a comprehensive survey of both types of registration, focusing on three-dimensional point clouds and meshes and 2) to provide a better understanding of registration from the perspective of data fitting. Registration is closely related to data fitting in which it comprises three core interwoven components: model selection, correspondences and constraints, and optimization. Study of these components 1) provides a basis for comparison of the novelties of different techniques, 2) reveals the similarity of rigid and nonrigid registration in terms of problem representations, and 3) shows how overfitting arises in nonrigid registration and the reasons for increasing interest in intrinsic techniques. We further summarize some practical issues of registration which include initializations and evaluations, and discuss some of our own observations, insights and foreseeable research trends.

In this paper we introduce a framework for the segmentation of scanned scrolled parchments, based on a novel graph cut based approach with an additional shape prior, in combination with anisotropic diffusion and geometry-constrained postprocessing. This problem has not been investigated by the computer vision community properly yet due to the parchment scanning technology novelty, and is extremely important for effective data recovery from historical scrolled documents whose content is inaccessible due to the deterioration of the parchment. To date, parchment segmentation has required user interaction, which is very time consuming for such data. We demonstrate with real examples how our algorithm is able to solve the major problem for scrolled parchment analysis, namely segment connected layers, and process the data without user interaction.

Designing rotational symmetry fields on surfaces is an important task for a wide range of graphics applications. This work introduces a rigorous and practical approach for automatic N-RoSy field design on arbitrary surfaces with user-defined field topologies. The user has full control of the number, positions, and indexes of the singularities (as long as they are compatible with necessary global constraints), the turning numbers of the loops, and is able to edit the field interactively. We formulate N-RoSy field construction as designing a Riemannian metric such that the holonomy along any loop is compatible with the local symmetry of N-RoSy fields. We prove the compatibility condition using discrete parallel transport. The complexity of N-RoSy field design is caused by curvatures. In our work, we propose to simplify the Riemannian metric to make it flat almost everywhere. This approach greatly simplifies the process and improves the flexibility such that it can design N-RoSy fields with single singularity and mixed-RoSy fields. This approach can also be generalized to construct regular remeshing on surfaces. To demonstrate the effectiveness of our approach, we apply our design system to pen-and-ink sketching and geometry remeshing. Furthermore, based on our remeshing results with high global symmetry, we generate Celtic knots on surfaces directly.

Principal curvatures and principal directions are fundamental local geometric properties. They are well defined on smooth surfaces. However, due to the nature as higher order differential quantities, they are known to be sensitive to noise. A recent work by Yang et al. combines principal component analysis with integral invariants and computes robust principal curvatures on multiple scales. Although the freedom of choosing the radius r gives results on different scales, in practice it is not an easy task to choose the most appropriate r for an arbitrary given model. Worse still, if the model contains features of different scales, a single r does not work well at all. In this work, we propose a scheme to automatically assign appropriate radii across the surface based on local surface characteristics. The radius r is not constant and adapts to the scale of local features. An efficient, iterative algorithm is used to approach the optimal assignment and the partition of unity is incorporated to smoothly combine the results with different radii. In this way, we can achieve a better balance between the robustness and the accuracy of feature locations. We demonstrate the effectiveness of our approach with robust principal direction field computation and feature extraction.

Developable surfaces have many desired properties in the manufacturing process. Since most existing CAD systems utilize tensor-product parametric surfaces including B-splines as design primitives, there is a great demand in industry to convert a general free-form parametric surface within a prescribed global error bound into developable patches. In this paper, we propose a practical and efficient solution to approximate a rectangular parametric surface with a small set ofC0-joint developable strips. The key contribution of the proposed algorithm is that, several optimization problems are elegantly solved in a sequence that offers a controllable global error bound on the developable surface approximation. Experimental results are presented to demonstrate the effectiveness and stability of the proposed algorithm.

Gradient mesh vector graphics representation, used in commercial software, is a regular grid with specified position and color, and their gradients, at each grid point. Gradient meshes can compactly represent smoothly changing data, and are typically used for single objects. This paper advances the state of the art for gradient meshes in several significant ways. Firstly, we introduce a topology-preserving gradient mesh representation which allows an arbitrary number of holes . This is important, as objects in images often have holes, either due to occlusion, or their 3D structure. Secondly, our algorithm uses the concept of image manifolds, adapting surface parameterization and fitting techniques to generate the gradient mesh in a fully automatic manner. Existing gradient-mesh algorithms require manual interaction to guide grid construction, and to cut objects with holes into disk-like regions. Our new algorithm is empirically at least 10 times faster than previous approaches. Furthermore, image segmentation can be used with our new algorithm to provide automatic gradient mesh generation for a whole image . Finally, fitting errors can be simply controlled to balance quality with storage.

3D mesh models are now widely available for use in various applications. The demand for automatic model analysis and understanding is ever increasing. Mesh segmentation is an important step towards model understanding, and acts as a useful tool for different mesh processing applications, e.g. reverse engineering and modeling by example. We extend a random walk method used previously for image segmentation to give algorithms for both interactive and automatic mesh segmentation. This method is extremely efficient, and scales almost linearly with increasing number of faces. For models of moderate size, interactive performance is achieved with commodity PCs. It is easy-to-implement, robust to noise in the mesh, and yields results suitable for downstream applications for both graphical and engineering models.

In this paper we present a new algorithm which turns an unstructured triangle mesh into a quad-dominant mesh with edges aligned to the principal directions of the underlying geometry. Instead of computing a globally smooth parameterization or integrating curvature lines along a tangent vector field, we simply apply an iterative relaxation scheme which incrementally aligns the mesh edges to the principal directions. The quad-dominant mesh is eventually obtained by dropping the not-aligned diagonals from the triangle mesh. A post-processing stage is introduced to further improve the results. The major advantage of our algorithm is its conceptual simplicity since it is merely based on elementary mesh operations such as edge collapse, flip, and split. The resulting meshes exhibit a very good alignment to surface features and rather uniform distribution of mesh vertices. This makes them very well-suited, e.g., as Catmull-Clark Subdivision control meshes.

Wireframe is a modeling tool widely used in industrial geometric design. The term wireframe refers to two sets of curves, with the property that each curve from one set intersects with each curve from the other set. Akin to the mu-, v-isocurves in a tensor-product surface, the two sets of curves in a wireframe span an underlying surface. In many industrial design activities, wireframes are usually set up and adjusted by the designers before the whole surfaces are reconstructed. For adjustment, the fairness of wireframe has a direct influence on the quality of the underlying surface. Wireframe fairing is significantly different from fairing individual curves in that intersections should be preserved and kept in the same order. In this paper, we first present a technique for wireframe fairing by fixing the parameters during fairing. The limitation of fixed parameters is further released by an iterative gradient descent optimization method with step-size control. Experimental results show that our solution is efficient, and produces reasonably fairing results of the wireframes.

Developable surfaces have many desired properties in manufacturing process. Since most existing CAD systems utilize parametric surfaces as the design primitive, there is a great demand in industry to convert a parametric surface within a prescribed global error bound into developable patches. In this work we propose a simple and efficient solution to approximate a general parametric surface with a minimum set of C0-joint developable strips. The key contribution of the proposed algorithm is that, several global optimization problems are elegantly solved in a sequence that offers a controllable global error bound on the developable surface approximation. Experimental results are presented to demonstrate the effectiveness and stability of the proposed algorithm.

Sharp edges, ridges, valleys, and prongs are critical for the appearance and an accurate representation of a 3D model. In this paper, we propose a novel approach that deals with the global shape of features in a robust way. Based on a remeshing algorithm which delivers an isotropic mesh in a feature-sensitive metric, features are recognized on multiple scales via integral invariants of local neighborhoods. Morphological and smoothing operations are then used for feature region extraction and classification into basic types such as ridges, valleys, and prongs. The resulting representation of feature regions is further used for feature-specific editing operations.

Segmenting meshes into natural regions is useful for model understanding and many practical applications. In this paper, we present a novel, automatic algorithm for segmenting meshes into meaningful pieces. Our approach is a clustering-based top-down hierarchical segmentation algorithm. We extend recent work on feature sensitive isotropic remeshing to generate a mesh hierarchy especially suitable for segmentation of large models with regions at multiple scales. Using integral invariants for estimation of local characteristics, our method is robust and efficient. Moreover, statistical quantities can be incorporated, allowing our approach to segment regions with different geometric characteristics or textures.

In this paper, we present an automatic method which can transfer geometric textures from one object to another, and can apply a manually designed geometric texture to a model. Our method is based on geometry images as introduced by Gu et al. The key ideas in this method involve geometric texture extraction, boundary consistent texture synthesis, discretized orientation and scaling, and reconstruction of synthesized geometry. Compared to other methods, our approach is efficient and easy-to-implement, and produces results of high quality.