4D head capture aims to generate dynamic facial meshes in the same topology with corresponding UV maps, which requires temporal correspondence between 3D head models. Existing pipelines either involve manual processing of artists or employ constraints such as landmark tracking and optical flow, failing to achieve a trade-off between accuracy and efficiency. To enhance this process, we propose Topo4D++, a novel framework for automatic geometry and texture reconstruction that optimizes densely aligned 4D heads and 8 K BRDF maps directly from calibrated multi-view videos. Our key insight is to represent facial models as a set of dynamic 3D Gaussians with fixed topology, where the Gaussian centers are bound to the mesh vertices. This enables tracking all vertices rather than sparse vertices on the face accurately by leveraging the inverse rendering capabilities of 3D Gaussian Splatting (3DGS), while also enabling ultra-high-resolution texture generation. To maintain face structure during dynamic 3DGS optimization, we propose to optimize geometry and texture alternatively under physical and topological constraints frame-by-frame and employ blendshape-based expression priors to address extreme expressions. Then, we propose to extract dynamic facial meshes in a regular wiring arrangement and high-fidelity textures with pore-level details from the learned Gaussians. Finally, we train a diffusion-based model to generate BRDF texture maps to achieve physically based rendering. Given the absence of a universal benchmark, we construct JHead, a novel benchmark for the comprehensive evaluation of 4D head capture methods. Extensive experiments on different datasets demonstrate that our method is generalized to different capture systems, identities, and expressions, outperforming current state-of-the-art head reconstruction methods in both mesh and texture qualitatively and quantitatively.

Appearance-based supervised methods with full-face image input have made tremendous advances in recent gaze estimation tasks. However, intensive human annotation requirement inhibits current methods from achieving industrial level accuracy and robustness. Although methods based on generative AI can synthesize eye images to expand self-annotated eye data, these methods usually have limited model capacity and cannot effectively inject gaze information, resulting in poor quality, monotonous texture, and inaccurate gaze direction of the generated eye images. To alleviate the above challenge, we propose a novel gaze data synthesizer framework, in which a 3D-eye model that can flexibly manipulate the gaze direction is used to finely control eye image synthesis based on a stable diffusion large generative model, so that high-quality eye images with arbitrary gaze angle can be synthesized. At the same time, when training the gaze feature extractor, we propose a cross-domain feature alignment module to minimize the feature distribution discrepancy between real samples and synthetic ones, to pursue domain-invariant (shape, texture, etc.) gaze representation. Both qualitative and quantitative experimental results demonstrate that the proposed scheme generates high-quality gaze images and also achieves superior gaze estimation performances over state-of-the-art.

3D Gaussian Splatting (3DGS) has attracted great attention in novel view synthesis because of its superior rendering efficiency and high fidelity. However, the trained Gaussians suffer from severe zooming degradation due to non-adjustable representation derived from single-scale training. Though some methods attempt to tackle this problem via post-processing techniques such as selective rendering or filtering techniques towards primitives, the scale-specific information is not involved in Gaussians. In this paper, we propose a unified optimization method to make Gaussians adaptive for arbitrary scales by self-adjusting the primitive properties (e.g., color, shape and size) and distribution (e.g., position). Inspired by the mipmap technique, we design pseudo ground-truth for the target scale and propose a scale-consistency guidance loss to inject scale information into 3D Gaussians. Our method is a plug-in module, applicable for any 3DGS models to solve the zoomin and zoom-out aliasing. Extensive experiments demonstrate the effectiveness of our method. Notably, our method outperforms 3DGS in PSNR by an average of 9.25 dB for zoom-in and 10.40 dB for zoom-out on NeRF Synthetic dataset. Our project website: https://github.com/renaissanceee/Mipmap-GS.

The imperative for compression of material textures emerges from the critical demand for high-quality rendering, which necessitates sophisticated textures that, in turn, require substantial storage and memory resources. Thus, low-bitrate compression is crucial, especially in modern games demanding higher texture resolutions. Concurrent methodologies in texture compression predominantly employ a block-based paradigm based on color space, which inevitably leads to representational redundancies and a limited compression scope, particularly at lower bitrates. In the context of mobile devices, bandwidth during texture loading and runtime memory are major bottlenecks, making existing compression algorithms inadequate for high-resolution textures. To mitigate these limitations, we propose a novel multi-resolution texture compression scheme, Neural Block Compression (NBC), developed within the neural feature domain. Our encoding scheme is constructed on a hierarchy of multi-resolution neural feature blocks, and the key ingredient is the variable bitrates quantization scheme. It allocates higher bitrates to higher feature mip-levels and lower bitrates to lower feature mip-levels, thereby extending the concept of block compression from color domain into neural feature domain. Extensive experiments demonstrate the superior texture compression quality achieved by the proposed scheme, especially at low bitrates.

We present InstantSticker, a disentangled reconstruction pipeline based on Image-Based Lighting (IBL), which focuses on highly realistic decal blending, simulates stickers attached to the reconstructed surface, and allows for instant editing and real-time rendering. To achieve stereoscopic impression of the decal, we introduce shadow factor into IBL, which can be adaptively optimized during training. This allows the shadow brightness of surfaces to be accurately decomposed rather than baked into the diffuse color, ensuring that the edited texture exhibits authentic shading. To address the issues of warping and blurriness in previous methods, we apply As-Rigid-As-Possible (ARAP) parameterization to pre-unfold a specified area of the mesh and use the local UV mapping combined with a neural texture map to enhance the ability to express high-frequency details in that area. For instant editing, we utilize the Disney BRDF model, explicitly defining material colors with 3-channel diffuse albedo. This enables instant replacement of albedo RGB values during the editing process, avoiding the prolonged optimization required in previous approaches. In our experiment, we introduce the Ratio Variance Warping (RVW) metric to evaluate the local geometric warping of the decal area. Extensive experimental results demonstrate that our method surpasses previous decal blending methods in terms of editing quality, editing speed and rendering speed, achieving the state-of-the-art.

Current parameterized image representations embed visual information along the semantic boundaries and struggle to express the internal detailed texture structures of image components, leading to a lack of content consistency after image editing and driving. To address these challenges, this work proposes a novel parameterized representation based on hierarchical image proxy geometry, utilizing multi-layer hierarchically interrelated proxy geometric control points to embed multi-scale long-range structures and fine-grained texture details. The proposed representation enables smoother and more continuous interpolation during image rendering and ensures high-quality consistency within image components during image editing. Additionally, under the layer-wise representation strategy based on semantic-aware image layer decomposition, we enable decoupled image shape/texture editing of the targets of interest within the image. Extensive experimental results on image vectorization and editing tasks demonstrate that our proposed method achieves high rendering accuracy of general images, including natural images, with a significantly higher image parameter compression ratio, facilitating user-friendly editing of image semantic components.

Geodesic distance serves as a reliable means of measuring distance in nonlinear spaces, and such nonlinear manifolds are prevalent in the current multimodal learning. In these scenarios, some samples may exhibit high similarity, yet they convey different semantics, making traditional distance metrics inadequate for distinguishing between positive and negative samples. This paper introduces geodesic distance as a novel distance metric in multi-modal learning for the first time, to mine correlations between samples, aiming to address the limitations of common distance metric. Our approach incorporates a comprehensive series of strategies to adapt geodesic distance for the current multi-modal learning. Specifically, we construct a graph structure to represent the adjacency relationships among samples by thresholding distances between them and then apply the shortest-path algorithm to obtain geodesic distance within this graph. To facilitate efficient computation, we further propose a hierarchical graph structure through clustering and combined with incremental update strategies for dynamic status updates. Extensive experiments across various downstream tasks validate the effectiveness of our proposed method, demonstrating its capability to capture complex relationships between samples and improve the performance of multimodal learning models.

Despite significant advances in accurately estimating geometry in contemporary single-image 3D human reconstruction, creating a high-quality, efficient, and animatable 3D avatar remains an open challenge. Two key obstacles persist: incomplete observation and inconsistent 3D priors. To address these challenges, we propose SinGS, aiming to achieve high-quality and efficient animatable 3D avatar reconstruction. At the heart of SinGS are two key components: Kinematic Human Diffusion and Geometry-Preserving 3D Gaussain Splatting. The former is a foundational human model that samples within pose space to generate a highly 3D-consistent and high-quality sequence of human images, inferring unseen viewpoints and providing kinematic priors. The latter is a system that reconstructs a compact, high-quality 3D avatar even under imperfect priors, achieved through a novel semantic Laplacian regularization and a geometry-preserving density control strategy that enable precise and compact assembly of 3D primitives. Extensive experiments demonstrate that SinGS enables lifelike, animatable human reconstructions, maintaining both high quality and inference efficiency (up to 70FPS).

Accurately and efficiently simulating complex fluid dynamics is a challenging task that has traditionally relied on computationally intensive methods. Neural network-based approaches, such as convolutional and graph neural networks, have partially alleviated this burden by enabling efficient local feature extraction. However, they struggle to capture long-range dependencies due to limited receptive fields, and Transformer-based models, while providing global context, incur prohibitive computational costs. To tackle these challenges, we propose AMR-Transformer, an efficient and accurate neural CFD-solving pipeline that integrates a novel adaptive mesh refinement scheme with a Navier-Stokes constraint-aware fast pruning module. This design encourages long-range interactions between simulation cells and facilitates the modeling of global fluid wave patterns, such as turbulence and shockwaves. Experiments show that our approach achieves significant gains in efficiency while preserving critical details, making it suitable for high-resolution physical simulations with long-range dependencies. On CFDBench, PDEBench and a new shock wave dataset, our pipeline demonstrates up to an order-of- magnitude improvement in accuracy over baseline models. Additionally, compared to ViT, our approach achieves a reduction in FLOPs of up to 60 times.

Scalable Vector Graphics (SVG) is a code structure used to represent visual information, and with the powerful capabilities of large language models, it holds significant research potential. Current text-to-SVG generation methods lack generalization capabilities and struggle with accurately adhering to input generation instructions. In this paper, we propose a novel approach for generating SVG using large language models, named SVGThinker, which incorporates a reasoning process to align the generation of SVG code with the visualization process, while supporting all SVG primitives. Through sequential rendering of SVG primitives, we first use a multimodal model to annotate the SVG, followed by sequential updates corresponding to the incremental additions of primitives. We then employ a supervised training framework based on Chain-of-Thought reasoning, which enhances the model's robustness and reduces the risk of errors or hallucinations. Through comparisons with state-of-the-art baseline models, our experiments show that our model generates more stable, high-quality, and editable SVG code. In contrast to image-based methods, our approach preserves the structural advantages of SVG and supports precise, hierarchical editing. We believe our work opens new directions for SVG generation, with potential applications in design, content creation, and automated SVG-based graphic generation.

We introduce the AR2 O Painter, an interactive intelligent system designed for real-time, highly realistic oil painting creation. This Agent can faithfully reproduce any portrait image, allowing users to visually enjoy the stroke-by-stroke painting process immersively as the artwork is completed within two minutes. It consists of two modules: the Oil Painting Stroke Sequence Planner, which performs multi-level semantic-based brushstroke sequence decomposition on portrait images, mimicking the logic of artist painting, and the Oil Painting Rendering Engine, which receives the brushstroke sequence, models the pigment via fluid dynamics, simulates its interaction with the canvas and brush, and applies a tailored PBR model with microfacet BRDF, Fresnel effects, and stroke-level geometry, enabling perceptually plausible gloss and fine-grained surface relief. To the best of our knowledge, it is the first real-time intelligent painting system to generate realistic oil paintings with high interactivity and artistic fidelity. The demo video is available at https://youtu.be/aN-W06GmnP8.

Virtual footwear try-on (VFTON), a critical yet underexplored area in virtual try-on (VTON), aims to synthesize faithful try-on results given diverse footwear and model images while maintaining 3D consistency and texture authenticity. Unlike conventional garment-focused VTON methods, VFTON presents unique challenges due to (1) Data Scarcity, which arises from the difficulty of perfectly matching product shoes with models wearing the identical ones, (2) Viewpoint Misalignment, where the target foot pose and source shoe views are always misaligned, leading to incomplete texture information and detail distortion, and (3) Background-induced Color Distortion, where complex material of footwear interacts with environmental lighting, causing unintended color contamination. To address these challenges, we introduce MVShoes, a multi-view shoe try-on dataset consisting of 7305 well-annotated image triplets, covering diverse footwear categories and challenging try-on scenarios. Furthermore, we propose a dual-stream DiT architecture, ShoeFit, designed to mitigate viewpoint misalignment through Multi-View Conditioning with 3D Rotary Position Embedding, and alleviate background-induced distortion using the LayeredRefAttention which leverages background features to modulate footwear latents. The proposed framework effectively decouples shoe appearance from environmental interferences while preserving high-quality texture detail through decoupled denoising and conditioning branches. Extensive quantitative and qualitative experiments demonstrate that our method substantially improves rendering fidelity and robustness under challenging real-world product shoes, establishing a new benchmark in high-fidelity footwear try-on synthesis. The dataset and benchmark will be publicly available upon acceptance of the paper.

We present PhysDiff-VTON, a diffusion-based framework for image-based virtual try-on that systematically addresses the dual challenges of garment deformation modeling and high-frequency detail preservation. The core innovation lies in integrating physics-inspired mechanisms into the diffusion process: a pose-guided deformable warping module simulates fabric dynamics by predicting spatial offsets conditioned on human pose semantics, while wavelet-enhanced feature decomposition explicitly preserves texture fidelity through frequency-aware attention. Further enhancing generation quality, a novel sampling strategy optimizes the denoising trajectory via least action principles, enforcing temporal coherence, spatial smoothness, and multi-scale structural consistency. Comprehensive evaluations across multiple datasets demonstrate significant improvements in both geometric plausibility and perceptual quality compared to existing approaches. The framework establishes a new paradigm for synthesizing photorealistic try-on images that adhere to physical constraints while maintaining intricate garment details, advancing the practical applicability of diffusion models in fashion technology.

Animating quadruped 3D objects, such as chairs and tables, typically involves three steps in the traditional computer graphics pipeline: Rigging, Skinning, and Retargeting. Commonly, prevailing methods for each specific step are conceived in isolation. For rigging and skinning steps, optimization-based methods are typically used, but these approaches tend to be slow and susceptible to variations in 3D mesh surfaces. For the retargeting step, the obtained results often fall short of expectations, especially when dealing with dissimilar source and target skeletons, leading to issues like joint twisting. The devised procedure is also time-intensive, resulting in a complex final pipeline. To this end, we present a unified framework, termed Mesh2Animation, providing an end-to-end solution to these challenges. In Mesh2Animation, a learning-based method is proposed for quadruped 3D skeleton estimation. We introduce both skeleton-level and mesh-level loss, allowing the rigging, skinning, and retargeting steps to be optimized simultaneously. Specifically, a general predicted estimation from the rigging step initializes the skeleton, making the skinning step faster and more accurate, which in turn leads to better results in the retargeting step. Finally, the rigging, skinning and retargeting processes are optimized simultaneously under static and temporal constraints. Additionally, we can construct a novel animating dataset termed ShapeNet2Animation (SN2Animation) based on the proposed method, which shows potential application for pose transfer. Qualitative and quantitative results on SN2Animation, ShapeNet, Object3D and ModelNet10 datasets for animation demonstrate that our method achieves competitive performance and shows promising generalization ability on quadruped 3D objects. Our project is available athttps://sites.google.com/view/mesh2animation.

Efficient and highly accurate lightweight gaze estimation method has been receiving increasing research attention due to the emergence of mobile interactive platforms such as mobile device and AR/VR. State-of-the-art deep learning based gaze estimation models suffer from either heavy computational architecture which is infeasible for mobile deployment or limited generalization capability which cannot deal with large diversity in eye texture or distinguish subtle/frequent pupil movement. To mitigate the above challenges, we propose a novel lightweight network structure featuring a deformable approximate large kernel which can effectively extend the receptive field to handle complicated eye movement and highly varying eye/gaze region appearance with very tight computational budget. In the meantime, we embed the training of the gaze estimator into a control information extraction module, which serves as a gaze-parameter input that modularizes a large generative model (Stable Diffusion V1.5) to output gaze-specific eye images. In this way, the great generalization capability of large generative model could be implicitly distilled/pursued into our lightweight gaze model. Extensive comparisons with various state-of-the-art gaze estimation methods demonstrate the superiority of our proposed model and training scheme in terms of both accuracy and model complexity.

Rib fractures are a common and potentially severe injury that can be challenging and labor-intensive to detect in CT scans. While there have been efforts to address this field, the lack of large-scale annotated datasets and evaluation benchmarks has hindered the development and validation of deep learning algorithms. To address this issue, the RibFrac Challenge was introduced, providing a benchmark dataset of over 5,000 rib fractures from 660 CT scans, with voxel-level instance mask annotations and diagnosis labels for four clinical categories (buckle, nondisplaced, displaced, or segmental). The challenge includes two tracks: a detection (instance segmentation) track evaluated by an FROC-style metric and a classification track evaluated by an F1-style metric. During the MICCAI 2020 challenge period, 243 results were evaluated, and seven teams were invited to participate in the challenge summary. The analysis revealed that several top rib fracture detection solutions achieved performance comparable or even better than human experts. Nevertheless, the current rib fracture classification solutions are hardly clinically applicable, which can be an interesting area in the future. As an active benchmark and research resource, the data and online evaluation of the RibFrac Challenge are available at the challenge website (https://ribfrac.grand-challenge.org/). In addition, we further analyzed the impact of two post-challenge advancements-large-scale pretraining and rib segmentation-based on our internal baseline for rib fracture detection. These findings lay a foundation for future research and development in AI-assisted rib fracture diagnosis.

Depth map super-resolution (DSR) plays an indispensable role in 3D vision. We discover an non-trivial spectral phenomenon: the components of high-resolution (HR) and low-resolution (LR) depth maps manifest the same intrinsic phase, and the spectral phase of RGB is a superset of them, which suggests that a phase-aware filter can assist in the precise use of RGB cues. Motivated by this, we propose an intrinsic phase-preserving DSR paradigm, named IPPNet, to fully exploit inter-modality collaboration in a mutually guided way. In a nutshell, a novel Phase-Preserving Filtering Module (PPFM) is developed to generate dynamic phase-aware filters according to the LR depth flow to filter out erroneous noisy components contained in RGB and then conduct depth enhancement via the modulation of the phase-preserved RGB signal. By stacking multiple PPFM blocks, the proposed IPPNet is capable of reaching a highly competitive restoration performance. Extensive experiments on various benchmark datasets, e.g., NYU v2, RGB-D-D, reach SOTA performance and also well demonstrate the validity of the proposed phase-preserving scheme. Code: https://github.com/neuralchen/IPPNet/.

While text-3D editing has made significant strides in leveraging score distillation sampling, emerging approaches still fall short in delivering separable, precise and consistent outcomes that are vital to content creation. In response, we introduce FocalDreamer, a framework that merges base shape with editable parts according to text prompts for fine-grained editing within desired regions. Specifically, equipped with geometry union and dual-path rendering, FocalDreamer assembles independent 3D parts into a complete object, tailored for convenient instance reuse and part-wise control. We propose geometric focal loss and style consistency regularization, which encourage focal fusion and congruent overall appearance. Furthermore, FocalDreamer generates high-fidelity geometry and PBR textures which are compatible with widely-used graphics engines. Extensive experiments have highlighted the superior editing capabilities of FocalDreamer in both quantitative and qualitative evaluations.

We develop a novel vectorized image representation scheme accommodating both shape/geometry and texture in a decoupled way, particularly tailored for reconstruction and editing tasks of artistic/design images such as Emojis and Cliparts. In the heart of this representation is a set of sparsely and unevenly located 2D control points. On one hand, these points constitute a collection of paramet-ric/vectorized geometric primitives (e.g., curves and closed shapes) describing the shape characteristics of the target image. On the other hand, local texture codes, in terms of implicit neural network parameters, are spatially dis-tributed into each control point, yielding local coordinate-to-RGB mappings within the anchored region of each con-trol point. In the meantime, a zero-shot learning algorithm is developed to decompose an arbitrary raster image into the above representation, for the sake of high-fidelity im-age vectorization with convenient editing ability. Extensive experiments on a series of image vectorization and editing tasks well demonstrate the high accuracy offered by our proposed method, with a significantly higher image com-pression ratio over prior art.

We propose a neural reflectance model (NeuBRDF) that offers highly versatile material representation, yet with light memory and neural computation consumption towards achieving real-time rendering. The results depicted in Fig. 1, rendered at full HD resolution on a contemporary desktop machine, demonstrate that our system achieves real-time performance with a wide variety of appearances, which is approached by the following two designs. Firstly, recognizing that the bidirectional reflectance is distributed in a sparse high-dimensional space, we propose to project the BRDF into two low-dimensional components, i.e. two hemisphere feature-grids for incoming and outgoing directions, respectively. Secondly, we distribute learnable neural reflectance primitives on our highly-tailored spherical surface grid. These primitives offer informative features for each hemisphere component and reduce the complexity of the feature learning network, leading to fast evaluation. These primitives are centrally stored in a codebook and can be shared across multiple grids and even across materials, based on low-cost indices stored in material-specific spher-ical surface grids. Our NeuBRDF, agnostic to the material, provides a unified framework for representing a variety of materials consistently. Comprehensive experimental results on measured BRDF compression, Monte Carlo simulated BRDF acceleration, and extension to spatially varying effects demonstrate the superior quality and generalizability achieved by the proposed scheme.

Micro-mesh (μ-mesh.) is a new graphics primitive for compact representation of extreme geometry, consisting of a low-polygon base mesh enriched by per micro-vertex displacement. A new generation of GPUs supports this structure with hardware evolution on μ-mesh ray tracing, achieving real-time rendering in pixel level geometric details. In this article, we present a differentiable framework to convert standard meshes into this efficient format, offering a holistic scheme in contrast to the previous stage-based methods. In our construction context, a μ-mesh is defined where each base triangle is a parametric primitive, which is then reparameterized with Laplacian operators for efficient geometry optimization. Our framework offers numerous advantages for high-quality μ-mesh production: (i) end-to-end geometry optimization and displacement baking; (ii) enabling the differentiation of renderings with respect to μ-mesh for faithful reprojectability; (iii) high scalability for integrating useful features for μ-mesh production and rendering, such as minimizing shell volume, maintaining the isotropy of the base mesh, and visual-guided adaptive level of detail. Extensive experiments on μ-mesh construction for a large set of high-resolution meshes demonstrate the superior quality achieved by the proposed scheme.

Intelligent generation of vector graphics has very promising applications in the fields of advertising and logo design, artistic painting, animation production, etc. However, current mainstream vector image generation methods lack the encoding of image appearance information that is associated with the original vector representation and therefore lose valid supervision signal from the strong correlation between the discrete vector parameter (drawing in-struction) sequence and the target shape/structure of the corresponding pixel image. On the one hand, the gener-ation process based on pure vector domain completely ignores the similarity measurement between shape parameter (and their combination) and the paired pixel image appearance pattern; on the other hand, two-stage methods (i.e., generation-and-vectorization) based on pixel diffusion followed by differentiable image-to-vector translation suf-fer from wrong error-correction signal caused by approxi-mate gradients. To address the above issues, we propose a novel generation framework based on dual-domain (vector-pixel) diffusion with cross-modality impulse signals from each other. First, in each diffusion step, the current representation extracted from the other domain is used as a condition variable to constrain the subsequent sampling operation, yielding shape-aware new parameterizations; second, independent supervision signals from both domains avoid the gradient error accumulation problem caused by cross-domain representation conversion. Extensive experimental results on popular benchmarks including font and icon datasets demonstrate the great advantages of our proposed framework in terms of generated shape quality.

With the rise of new 3D representations like NeRF and 3D Gaussian splatting, creating realistic 3D scenes is easier than ever before. However, the incompatibility of these 3D representations with existing editing software has also introduced unprecedented challenges to 3D editing tasks. Although recent advances in text-to-image generative models have made some progress in 3D editing, these methods either lack precision or require users to manually specify the editing areas in 3D space, complicating the editing process. To overcome these issues, we propose Edit3D, an innovative 3D editing method designed to enhance editing quality. Specifically, we propose a multi-turn editing framework and introduce an attention-driven open-set segmentation (ADSS) technique within this framework. ADSS allows for more precise segmentation of parts, which enhances the editing precision and minimizes interference with pixels in areas that are not being edited. Additionally, we propose a fine-tuning phase, intended to further improve the overall editing quality without compromising the training efficiency. Experiments demonstrate that Edit3D effectively adjusts 3D scenes based on textual instructions. Through continuous and multiple turns of editing, it achieves more intricate combinations, enhancing the diversity of 3D editing effects. Code is available at https://github.com/PeterouZh/Edit3D.

While image-based virtual try-on has made significant strides, emerging approaches still fall short of delivering high-fidelity and robust fitting images across various scenarios, as their models suffer from issues of ill-fitted garment styles and quality degrading during the training process, not to mention the lack of support for various combinations of attire. Therefore, we first propose a lightweight, scalable, operator known as Hydra Block for attire combinations. This is achieved through a parallel attention mechanism that facilitates the feature injection of multiple garments from conditionally encoded branches into the main network. Secondly, to significantly enhance the model's robustness and expressiveness in real-world scenarios, we evolve its potential across diverse settings by synthesizing the residuals of multiple models, as well as implementing a mask region boost strategy to overcome the instability caused by information leakage in existing models. Equipped with the above design, AnyFit surpasses all baselines on high-resolution benchmarks and real-world data by a large gap, excelling in producing well-fitting garments replete with photorealistic and rich details. Furthermore, AnyFit’s impressive performance on high-fidelity virtual try-ons in any scenario from any image, paves a new path for future research within the fashion community.

Face identity editing (FIE) shows great value in AI content creation. Low-resolution FIE approaches have achieved tremendous progress, but high-quality FIE struggles. Two major challenges hinder higher-resolution and higher-performance development of FIE: lack of high-resolution dataset and unacceptable complexity forbidding for mobile platforms. To address both issues, we establish a novel large-scale, high-quality dataset tailored for FIE. Based on our SimSwap (Chen et al. 2020), we propose an upgraded version named SimSwap++ with significantly boosted model efficiency. SimSwap++ features two major innovations for high-performance model compression. First, a novel computational primitive named Conditional Dynamic Convolution (CD-Conv) is proposed to address the inefficiency of conditional schemes (e.g., AdaIN) in tiny models. CD-Conv achieves anisotropic processing and injection with significantly lower complexity compared to standard conditional operators, e.g., modulated convolution. Second, a Morphable Knowledge Distillation (MKD) is presented to further trim the overall model. Unlike conventional homogeneous teacher-student structures, MKD is designed to be heterogeneous and mutually compensable, endowing the student with the multi-path morphable property; thus, our student maximally inherits the teacher's knowledge after distillation while further reducing its complexity through structure re-parameterization. Extensive experiments demonstrate that our SimSwap++ achieves state-of-the-art performance (97.55% ID accuracy on FaceForensics++) with extremely low complexity (2.5 GFLOPs).

Multi-Source Domain Adaptation (MSDA) focuses on transferring the knowledge from multiple source domains to the target domain, which is a more practical and challenging problem compared to the conventional single-source domain adaptation. In this problem, it is essential to model multiple source domains and target domain jointly, and an effective domain combination scheme is also highly required. The graphical structure among different domains is useful to tackle these challenges, in which the interdependency among various instances/categories can be effectively modeled. In this work, we propose two types of graphical models, i.e. Conditional Random Field for MSDA (CRF-MSDA) and Markov Random Field for MSDA (MRF-MSDA), for cross-domain joint modeling and learnable domain combination. In a nutshell, given an observation set composed of a query sample and the semantic prototypes (i.e. representative category embeddings) on various domains, the CRF-MSDA model seeks to learn the joint distribution of labels conditioned on the observations. We attain this goal by constructing a relational graph over all observations and conducting local message passing on it. By comparison, MRF-MSDA aims to model the joint distribution of observations over different Markov networks via an energy-based formulation, and it can naturally perform label prediction by summing the joint likelihoods over several specific networks. Compared to the CRF-MSDA counterpart, the MRF-MSDA model is more expressive and possesses lower computational cost. We evaluate these two models on four standard benchmark data sets of MSDA with distinct domain shift and data complexity, and both models achieve superior performance over existing methods on all benchmarks. In addition, the analytical studies illustrate the effect of different model components and provide insights about how the cross-domain joint modeling performs.

Various methods have been proposed to defend against adversarial attacks. However, there is a lack of enough theoretical guarantee of the performance, thus leading to two problems: First, deficiency of necessary adversarial training samples might attenuate the normal gradient's back-propagation, which leads to overfitting and gradient masking potentially. Second, point-wise adversarial sampling offers an insufficient support region for adversarial data and thus cannot form a robust decision-boundary. To solve these issues, we provide a theoretical analysis to reveal the relationship between robust accuracy and the complexity of the training set in adversarial training. As a result, we propose a novel training scheme called Variational Adversarial Defense. Based on the distribution of adversarial samples, this novel construction upgrades the defend scheme from local point-wise to distribution-wise, yielding an enlarged support region for safeguarding robust training, thus possessing a higher promising to defense attacks. The proposed method features the following advantages: 1) Instead of seeking adversarial examples point-by-point (in a sequential way), we draw diverse adversarial examples from the inferred distribution; and 2) Augmenting the training set by a larger support region consolidates the smoothness of the decision boundary. Finally, the proposed method is analyzed via the Taylor expansion technique, which casts our solution with natural interpretability.

Automatic rib labeling and anatomical centerline extraction are common prerequisites for various clinical applications. Prior studies either use in-house datasets that are inaccessible to communities, or focus on rib segmentation that neglects the clinical significance of rib labeling. To address these issues, we extend our prior dataset (RibSeg) on the binary rib segmentation task to a comprehensive benchmark, named RibSeg v2, with 660 CT scans (15,466 individual ribs in total) and annotations manually inspected by experts for rib labeling and anatomical centerline extraction. Based on the RibSeg v2, we develop a pipeline including deep learning-based methods for rib labeling, and a skeletonization-based method for centerline extraction. To improve computational efficiency, we propose a sparse point cloud representation of CT scans and compare it with standard dense voxel grids. Moreover, we design and analyze evaluation metrics to address the key challenges of each task. Our dataset, code, and model are available online to facilitate open research at https://github.com/M3DV/RibSeg.

While orthokeratology (OK) has shown effective to slow the progression of myopia, it remains unknown how spatially distributed structural stress/tension applying to different regions affects the change of corneal geometry, and consecutive the outcome of myopia control, at fine-grained detail. Acknowledging that the underlying working mechanism of OK lens is essentially mechanics induced refractive parameter reshaping, in this study, we develop a novel mechanics rule guided deep image-to-image learning framework, which densely predicts patient's corneal topography change according to treatment parameters (lens geometry, wearing time, physiological parameters, etc.), and consecutively predicts the influence on eye axial length change after OK treatment. Encapsulated in a U-shaped multi-resolution map-to-map architecture, the proposed model features two major components. First, geometric and wearing parameters of OK lens are spatially encoded with convolutions to form a multi-channel input volume/tensor for latent encodings of external stress/tension applied to different regions of cornea. Second, these external latent force maps are progressively down-sampled and injected into this multi-scale architecture for predicting the change of corneal topography map. At each feature learning layer, we formally derive a mathematic framework that simulates the physical process of corneal deformation induced by lens-to-cornea interaction and corneal internal tension, which is reformulated into parameter learnable cross-attention/self-attention modules in the context of transformer architecture. A total of 1854 eyes of myopia patients are included in the study and the results show that the proposed model precisely predicts corneal topography change with a high PSNR as 28.45dB, as well as a significant accuracy gain for axial elongation prediction (i.e., 0.0276 in MSE). It is also demonstrated that our method provides interpretable associations between various OK treatment parameters and the final control effect. Our project code package is available at https://github.com/Rongdingyi/PhyIntNet.

Noninvasively and accurately predicting the epidermal growth factor receptor (EGFR) mutation status is a clinically vital problem. Moreover, further identifying the most suspicious area related to the EGFR mutation status can guide the biopsy to avoid false negatives. Deep learning methods based on computed tomography (CT) images may improve the noninvasive prediction of EGFR mutation status and potentially help clinicians guide biopsies by visual methods. Inspired by the potential inherent links between EGFR mutation status and invasiveness information, we hypothesized that the predictive performance of a deep learning network can be improved through extra utilization of the invasiveness information. Here, we created a novel explainable transformer network for EGFR classification named gated multiple instance learning transformer (GMILT) by integrating multi-instance learning and discriminative weakly supervised feature learning. Pathological invasiveness information was first introduced into the multitask model as embeddings. GMILT was trained and validated on a total of 512 patients with adenocarcinoma and tested on three datasets (the internal test dataset, the external test dataset, and The Cancer Imaging Archive (TCIA) public dataset). The performance (area under the curve (AUC) =0.772 on the internal test dataset) of GMILT exceeded that of previously published methods and radiomics-based methods (i.e., random forest and support vector machine) and attained a preferable generalization ability (AUC =0.856 in the TCIA test dataset and AUC =0.756 in the external dataset). A diameter-based subgroup analysis further verified the efficiency of our model (most of the AUCs exceeded 0.772) to noninvasively predict EGFR mutation status from computed tomography (CT) images. In addition, because our method also identified the "core area" of the most suspicious area related to the EGFR mutation status, it has the potential ability to guide biopsies.

Depth map super-resolution (DSR) has been a fundamental task for 3D computer vision. While arbitrary scale DSR is a more realistic setting in this scenario, previous approaches predominantly suffer from the issue of inefficient real-numbered scale upsampling. To explicitly address this issue, we propose a novel continuous depth representation for DSR. The heart of this representation is our proposed Geometric Spatial Aggregator (GSA), which exploits a distance field modulated by arbitrarily upsampled target gridding, through which the geometric information is explicitly introduced into feature aggregation and target generation. Furthermore, bricking with GSA, we present a transformer-style backbone named GeoDSR, which possesses a principled way to construct the functional mapping between local coordinates and the high-resolution output results, empowering our model with the advantage of arbitrary shape transformation ready to help diverse zooming demand. Extensive experimental results on standard depth map benchmarks, e.g., NYU v2, have demonstrated that the proposed framework achieves significant restoration gain in arbitrary scale depth map super-resolution compared with the prior art. Our codes are available at https://github.com/nana01219/GeoDSR.

Spatial audio, which focuses on immersive 3D sound rendering, is widely applied in the acoustic industry. One of the key problems of current spatial audio rendering methods is the lack of personalization based on different anatomies of individuals, which is essential to produce accurate sound source positions. In this work, we address this problem from an interdisciplinary perspective. The rendering of spatial audio is strongly correlated with the 3D shape of human bodies, particularly ears. To this end, we propose to achieve personalized spatial audio by reconstructing 3D human ears with single-view images. First, to benchmark the ear reconstruction task, we introduce AudioEar3D, a high-quality 3D ear dataset consisting of 112 point cloud ear scans with RGB images. To self-supervisedly train a reconstruction model, we further collect a 2D ear dataset composed of 2,000 images, each one with manual annotation of occlusion and 55 landmarks, named AudioEar2D. To our knowledge, both datasets have the largest scale and best quality of their kinds for public use. Further, we propose AudioEarM, a reconstruction method guided by a depth estimation network that is trained on synthetic data, with two loss functions tailored for ear data. Lastly, to fill the gap between the vision and acoustics community, we develop a pipeline to integrate the reconstructed ear mesh with an off-the-shelf 3D human body and simulate a personalized Head-Related Transfer Function (HRTF), which is the core of spatial audio rendering. Code and data are publicly available in https://github.com/seanywang0408/AudioEar.

Recent years we have witnessed rapid development in NeRF-based image rendering due to its high quality. However, point clouds rendering is somehow less explored. Compared to NeRF-based rendering which suffers from dense spatial sampling, point clouds rendering is naturally less computation intensive, which enables its deployment in mobile computing device. In this work, we focus on boosting the image quality of point clouds rendering with a compact model design. We first analyze the adaption of the volume rendering formulation on point clouds. Based on the analysis, we simplify the NeRF representation to a spatial mapping function which only requires single evaluation per pixel. Further, motivated by ray marching, we rectify the the noisy raw point clouds to the estimated intersection between rays and surfaces as queried coordinates, which could avoid spatial frequency collapse and neighbor point disturbance. Composed of rasterization, spatial mapping and the refinement stages, our method achieves the state-of-the-art performance on point clouds rendering, outperforming prior works by notable margins, with a smaller model size. We obtain a PSNR of 31.74 on NeRF-Synthetic, 25.88 on ScanNet and 30.81 on DTU. Code and data are publicly available in https://github.com/seanywang0408/RadianceMapping.

Recent works on learning-based frameworks for Lagrangian (i.e., particle-based) fluid simulation, though bypassing iterative pressure projection via efficient convolution operators, are still time-consuming due to excessive amount of particles. To address this challenge, we propose a dynamic multi-scale gridding method to reduce the magnitude of elements that have to be processed, by observing repeated particle motion patterns within certain consistent regions. Specifically, we hierarchically generate multi-scale micelles in Euclidean space by grouping particles that share similar motion patterns/characteristics based on super-light motion and scale estimation modules. With little internal motion variation, each micelle is modeled as a single rigid body with convolution only applied to a single representative particle. In addition, a distance-based interpolation is conducted to propagate relative motion message among micelles. With our efficient design, the network produces high visual fidelity fluid simulations with the inference time to be only 4.24 ms/frame (with 6K fluid particles), hence enables real-time human-computer interaction and animation. Experimental results on multiple datasets show that our work achieves great simulation acceleration with negligible prediction error increase.

In this paper, we identify symmetry property in adversarial scenario by viewing adversarial attack in a fine-grained manner. A newly designed metric called attack proportion, is thus proposed to count the proportion of the adversarial examples misclassified between classes. We observe that the distribution of attack proportion is unbalanced as each class shows vulnerability to particular classes. Further, some class pairs correlate strongly and have the same degree of attack proportion for each other. We call this intriguing phenomenon symmetry property. We empirically prove this phenomenon is widespread and then analyze the reason behind the existence of symmetry property. This explanation, to some extent, could be utilized to understand robust models, which also inspires us to strengthen adversarial defenses.

The ability of scale-equivariance processing blocks plays a central role in arbitrary-scale image super-resolution tasks. Inspired by this crucial observation, this work proposes two novel scale-equivariant modules within a transformer-style framework to enhance arbitrary-scale image super-resolution (ASISR) performance, especially in high upsampling rate image extrapolation. In the feature extraction phase, we design a plug-in module called Adaptive Feature Extractor, which injects explicit scale information in frequency-expanded encoding, thus achieving scale-adaption in representation learning. In the upsampling phase, a learnable Neural Kriging upsampling operator is introduced, which simultaneously encodes both relative distance (i.e., scale-aware) information as well as feature similarity (i.e., with priori learned from training data) in a bilateral manner, providing scale-encoded spatial feature fusion. The above operators are easily plugged into multiple stages of a SR network, and a recent emerging pretraining strategy is also adopted to impulse the model's performance further. Extensive experimental results have demonstrated the outstanding scale-equivariance capability offered by the proposed operators and our learning framework, with much better results than previous SOTA methods at arbitrary scales for SR. Our code is available at https://github.com/neura1chen/EQSR

Learning continuous image representations is recently gaining popularity for image super-resolution (SR) because of its ability to reconstruct high-resolution images with arbitrary scales from low-resolution inputs. Existing methods mostly ensemble nearby features to predict the new pixel at any queried coordinate in the SR image. Such a local ensemble suffers from some limitations: i) it has no learnable parameters and it neglects the similarity of the visual features; ii) it has a limited receptive field and cannot ensemble relevant features in a large field which are important in an image. To address these issues, this paper proposes a continuous implicit attention-in-attention network, called CiaoSR. We explicitly design an implicit attention network to learn the ensemble weights for the nearby local features. Furthermore, we embed a scale-aware attention in this implicit attention network to exploit additional non-local information. Extensive experiments on benchmark datasets demonstrate CiaoSR significantly outperforms the existing single image SR methods with the same backbone. In addition, CiaoSR also achieves the state-of-the-art performance on the arbitrary-scale SR task. The effectiveness of the method is also demonstrated on the real-world SR setting. More importantly, CiaoSR can be flexibly integrated into any backbone to improve the SR performance.

We develop a simple yet surprisingly effective implicit representing scheme called Multiplicative Fourier Level of Detail (MFLOD) motivated by the recent success of multiplicative filter network. Built on multi-resolution feature grid/volume (e.g., the sparse voxel octree), each level's feature is first modulated by a sinusoidal function and then element-wisely multiplied by a linear transformation of previous layer's representation in a layer-to-layer recursive manner, yielding the scale-aggregated encodings for a subsequent simple linear forward to get final output. In contrast to previous hybrid representations relying on interleaved multilevel fusion and nonlinear activation-based decoding, MFLOD could be elegantly characterized as a linear combination of sine basis functions with varying amplitude, frequency, and phase upon the learned multilevel features, thus offering great feasibility in Fourier analysis. Comprehensive experimental results on implicit neural representation learning tasks including image fitting, 3D shape representation, and neural radiance fields well demonstrate the superior quality and generalizability achieved by the proposed MFLOD scheme.

We develop a generalized 3D shape generation prior model, tailored for multiple 3D tasks including unconditional shape generation, point cloud completion, and cross-modality shape generation, etc. On one hand, to precisely capture local fine detailed shape information, a vector quantized variational autoencoder (VQ-VAE) is utilized to index local geometry from a compactly learned code-book based on a broad set of task training data. On the other hand, a discrete diffusion generator is introduced to model the inherent structural dependencies among different tokens. In the meantime, a multi-frequency fusion module (MFM) is developed to suppress high-frequency shape feature fluctuations, guided by multi-frequency contextual information. The above designs jointly equip our proposed 3D shape prior model with high-fidelity, diverse features as well as the capability of cross-modality alignment, and extensive experiments have demonstrated superior performances on various 3D shape generation tasks.

In this paper, we identify pattern imbalance from several aspects, and further develop a new training scheme to avert pattern preference as well as spurious correlation. In contrast to prior methods which are mostly concerned with category or domain granularity, ignoring the potential finer structure that existed in datasets, we give a new definition of seed category as an appropriate optimization unit to distinguish different patterns in the same category or domain. Extensive experiments on domain generalization datasets of diverse scales demonstrate the effectiveness of the proposed method.

While lightweight ViT framework has made tremendous progress in image super-resolution, its uni-dimensional self-attention modeling, as well as homogeneous aggregation scheme, limit its effective receptive field (ERF) to include more comprehensive interactions from both spatial and channel dimensions. To tackle these drawbacks, this work proposes two enhanced components under a new Omni-SR architecture. First, an Omni Self-Attention (OSA) block is proposed based on dense interaction principle, which can simultaneously model pixel-interaction from both spatial and channel dimensions, mining the potential correlations across omni-axis (i.e., spatial and channel). Coupling with mainstream window partitioning strategies, OSA can achieve superior performance with compelling computational budgets. Second, a multi-scale interaction scheme is proposed to mitigate sub-optimal ERF (i.e., premature saturation) in shallow models, which facilitates local propagation and meso-/global-scale interactions, rendering an omni-scale aggregation building block. Extensive experiments demonstrate that Omni-SR achieves recordhigh performance on lightweight super-resolution benchmarks (e.g., 26.95dB@Urban100 x4 with only 792K parameters). Our code is available at https://github.com/Francis0625/Omni-SR.

We develop an effective point cloud rendering pipeline for novel view synthesis, which enables high fidelity local detail reconstruction, real-time rendering and user-friendly editing. In the heart of our pipeline is an adaptive frequency modulation module called Adaptive Frequency Net (AFNet), which utilizes a hypernetwork to learn the local texture frequency encoding that is consecutively injected into adaptive frequency activation layers to modulate the implicit radiance signal. This mechanism improves the frequency expressive ability of the network with richer frequency basis support, only at a small computational budget. To further boost performance, a preprocessing module is also proposed for point cloud geometry optimization via point opacity estimation. In contrast to implicit rendering, our pipeline supports high-fidelity interactive editing based on point cloud manipulation. Extensive experimental results on NeRF-Synthetic, ScanNet, DTU and Tanks and Temples datasets demonstrate the superior performances achieved by our method in terms of PSNR, SSIM and LPIPS, in comparison to the state-of-the-art. Code is released at https://github.com/yizhangphd/FreqPCR.

Person search (PS) aims to jointly localize and identify a query person from natural, uncropped images. Existing works unintentionally adopt pedestrians (with similar poses and unchanging clothing) as the query and restrict the application scenarios in surveillance. This is due to that most PS datasets are collected from surveillance cameras with a limited diversity of views, scenes, appearances, etc. In this paper, we study a more general and realistic task in the wild, where we aim to search target persons with a much higher degree of diversity. To this end, we introduce a new PS dataset, namely MovieNet-PS, based on an existing large-scale movie dataset. MovieNet-PS is currently the largest and most diverse PS dataset, consisting of 160K images (100K for training), 274K bounding boxes, and 3K identities. It stands out from existing counterparts from two levels of diversities, i.e., scene-level and identity-level, with 92,043 scenes and significant variations in poses, clothing, scales, etc. for the same identity. To validate the rich context information on our dataset and make full use of it, we propose a novel global-local context network, which exploits scene and group context to boost the search performance. Extensive experiments demonstrate that MovieNet-PS is more challenging and comprehensive than existing datasets, and our approach further pushes the state of the art by a large margin (relatively 34% in mAP) on this dataset. Codes, models, and the dataset are available at: https://github.com/ZhengPeng7/GLCNet.

This work explores a novel image geometric abstraction paradigm based on assembly out of a pool of pre-defined simple parametric primitives (i.e., triangle, rectangle, circle and semicircle), facilitating controllable shape editing in images. While cast as a mixed combinatorial and continuous optimization problem, the above task is approximately reformulated within a token translation neural framework that simultaneously outputs primitive assignments and corresponding transformation and color parameters in an image-to-set manner, thus bypassing complex/non-differentiable graph-matching iterations. To relax the searching space and address the vanishing gradient issue, a novel Neural Soft Assignment scheme that well explores the quasi-equivalence between the assignment in Bipartite b-Matching and opacity-aware weighted multiple rasterization combination is introduced, drastically reducing the optimization complexity. Without ground-truth image abstraction labeling (i.e., vectorized representation), the whole pipeline is end-to-end trainable in a self-supervised manner, based on the linkage of differentiable rasterization techniques. Extensive experiments on several datasets well demonstrate that our framework is able to predict highly compelling vectorized geometric abstraction results with a combination of ONLY four simple primitives, also with VERY straightforward shape editing capability by simple replacement of primitive type, compared to previous image abstraction and image vectorization methods.

Shape primitives decomposition has been an important and long-standing task in 3D shape analysis. Prior arts heavily rely on 3D point clouds or voxel data for shape primitives extraction, which are less practical in real-world scenarios. This paper proposes to learn shape primitives from multi-view images by introducing implicit surface rendering. It is challenging since implicit shapes have a high degree of freedom, which violates the simplicity property of shape primitives. In this work, a novel regularization term named Implicit Convexity Regularization (ICR) imposed on implicit primitive learning is proposed to tackle this problem. We start with the convexity definition of general 3D shapes, and then derive the equivalent expression for implicit shapes represented by signed distance functions (SDFs). Further, instead of directly constraining the output SDF values which cause unstable optimization, we alternatively impose constraint on second order directional derivatives on line segments inside the shapes, which proves to be a tighter condition for 3D convexity. Implicit primitives constrained by the proposed ICR are combined into a whole object via softmax-weighted-sum operation over all primitive SDFs. Experiments on synthetic and real-world datasets show that our method is able to decompose objects into simple and reasonable shape primitives without the need of segmentation labels or 3D data. Code and data is publicly available in https://github.com/seanywang0408/ICR.

We consider a new problem of adapting a human mesh reconstruction model to out-of-domain streaming videos, where the performance of existing SMPL-based models is significantly affected by the distribution shift represented by different camera parameters, bone lengths, backgrounds, and occlusions. We tackle this problem through online adaptation, gradually correcting the model bias during testing. There are two main challenges: First, the lack of 3D annotations increases the training difficulty and results in 3D ambiguities. Second, non-stationary data distribution makes it difficult to strike a balance between fitting regular frames and hard samples with severe occlusions or dramatic changes. To this end, we propose the Dynamic Bilevel Online Adaptation algorithm (DynaBOA). It first introduces the temporal constraints to compensate for the unavailable 3D annotations and leverages a bilevel optimization procedure to address the conflicts between multi-objectives. DynaBOA provides additional 3D guidance by co-training with similar source examples retrieved efficiently despite the distribution shift. Furthermore, it can adaptively adjust the number of optimization steps on individual frames to fully fit hard samples and avoid overfitting regular frames. DynaBOA achieves state-of-the-art results on three out-of-domain human mesh reconstruction benchmarks.

The key point for an experienced craftsman to repair broken objects effectively is that he must know about them deeply. Similarly, we believe that a model can capture rich geometry information from a shape/scene and generate discriminative representations if it is able to find distorted parts of shapes/scenes and restore them. Inspired by this observation, we propose a novel self-supervised 3D learning paradigm named learning by restoring broken shapes/scenes (collectively called 3D geometry). We first develop a destroy-method cluster, from which we sample methods to break some local parts of an object. Then the destroyed object and the normal object are both sent into a point cloud network to get representations, which are employed to segment points that belong to distorted parts and further reconstruct/restore them to normal. To perform better in these two associated pretext tasks, the model is constrained to capture useful object features, such as rich geometric and contextual information. The object representations learned by this self-supervised paradigm transfer well to different datasets and perform well on downstream classification, segmentation and detection tasks. Experimental results on shape datasets and scene datasets demonstrate that our method achieves state-of-the-art performance among unsupervised methods. We also show experimentally that pre-training with our framework significantly boosts the performance of supervised models.

Capturing the interactions of human articulations lies in the center of skeleton-based action recognition. Recent graph-based methods are inherently limited in the weak spatial context modeling capability due to fixed interaction pattern and inflexible shared weights of GCN. To address above problems, we propose the multi-view interactional graph network (MV-IGNet) which can construct, learn and infer multi-level spatial skeleton context, including view-level (global), group-level, joint-level (local) context, in a unified way. MV-IGNet leverages different skeleton topologies as multi-views to cooperatively generate complementary action features. For each view, separable parametric graph convolution (SPG-Conv) enables multiple parameterized graphs to enrich local interaction patterns, which provides strong graph-adaption ability to handle irregular skeleton topologies. We also partition the skeleton into several groups and then the higher-level group contexts including inter-group and intra-group, are hierarchically captured by above SPG-Conv layers. A simple yet effective global context adaption (GCA) module facilitates representative feature extraction by learning the input-dependent skeleton topologies. Compared to the mainstream works, MV-IGNet can be readily implemented while with smaller model size and faster inference. Experimental results show the proposed MV-IGNet achieves impressive performance on large-scale benchmarks: NTU-RGB+D and NTU-RGB+D 120.

Learning to re-identify or retrieve a group of people across non-overlapped camera systems has important applications in video surveillance. However, most existing methods focus on (single) person re-identification (re-id), ignoring the fact that people often walk in groups in real scenarios. In this work, we take a step further and consider employing context information for identifying groups of people, i.e., group re-id. On the one hand, group re-id is more challenging than single person re-id, since it requires both a robust modeling of local individual person appearance (with different illumination conditions, pose/viewpoint variations, and occlusions), as well as full awareness of global group structures (with group layout and group member variations). On the other hand, we believe that person re-id can be greatly enhanced by incorporating additional visual context from neighboring group members, a task which we formulate as group-aware (single) person re-id. In this paper, we propose a novel unified framework based on graph neural networks to simultaneously address the above two group-based re-id tasks, i.e., group re-id and group-aware person re-id. Specifically, we construct a context graph with group members as its nodes to exploit dependencies among different people. A multi-level attention mechanism is developed to formulate both intra-group and inter-group context, with an additional self-attention module for robust graph-level representations by attentively aggregating node-level features. The proposed model can be directly generalized to tackle group-aware person re-id using node-level representations. Meanwhile, to facilitate the deployment of deep learning models on these tasks, we build a new group re-id dataset which contains more than 3.8K images with 1.5K annotated groups, an order of magnitude larger than existing group re-id datasets. Extensive experiments on the novel dataset as well as three existing datasets clearly demonstrate the effectiveness of the proposed framework for both group-based re-id tasks.

Style-based GANs achieve state-of-the-art results for generating high-quality images, but lack explicit and precise control over camera poses. Recently proposed NeRF-based GANs have made great progress towards 3D-aware image generation. However, the methods either rely on convolution operators which are not rotationally invariant, or utilize complex yet suboptimal training procedures to integrate both NeRF and CNN sub-structures, yielding un-robust, low-quality images with a large computational burden. This article presents an upgraded version called CIPS-3D++, aiming at high-robust, high-resolution and high-efficiency 3D-aware GANs. On the one hand, our basic model CIPS-3D, encapsulated in a style-based architecture, features a shallow NeRF-based 3D shape encoder as well as a deep MLP-based 2D image decoder, achieving robust image generation/editing with rotation-invariance. On the other hand, our proposed CIPS-3D++, inheriting the rotational invariance of CIPS-3D, together with geometric regularization and upsampling operations, encourages high-resolution high-quality image generation/editing with great computational efficiency. Trained on raw single-view images, without any bells and whistles, CIPS-3D++ sets new records for 3D-aware image synthesis, with an impressive FID of 3.2 on FFHQ at the 1024×1024 resolution. In the meantime, CIPS-3D++ runs efficiently and enjoys a low GPU memory footprint so that it can be trained end-to-end on high-resolution images directly, in contrast to previous alternate/progressive methods. Based on the infrastructure of CIPS-3D++, we propose a 3D-aware GAN inversion algorithm named FlipInversion, which can reconstruct the 3D object from a single-view image. We also provide a 3D-aware stylization method for real images based on CIPS-3D++ and FlipInversion. In addition, we analyze the problem of mirror symmetry suffered in training, and solve it by introducing an auxiliary discriminator for the NeRF network. Overall, CIPS-3D++ provides a strong base model that can serve as a testbed for transferring GAN-based image editing methods from 2D to 3D.

Supervised learning is a mainstay for large discriminative models in 3D computer vision, while large amounts of human-annotated data are the key to achieve state-of-the-art performance. This limitation is particularly notable for large-scale point cloud sequence segmentation tasks, because point-level annotations are very time-consuming and especially expensive. To overcome this challenge, we develop a novel semi-supervised framework for point cloud sequences segmentation. Specifically, we develop two kinds of pseudo labeling methods with extracting global semantic information from labeled frames and dynamic information from each sequence respectively. Then the two kinds of generated labels are combined as more robust pseudo labels (GD-Pseudo labels) for unlabeled frames. We finally apply an efficient iterative learning scheme to train a model with a small quantity of human-annotated data and large-scale pseudo-labeled data. Equipped with our framework, the model achieves significant performance improvement (+12—25 mIoU) on SemanticKITTI and Synthia when compared with frameworks that do not utilize large amounts of unlabeled data. Moreover, our method achieves comparable performance with only 20% annotated frames on SemanticKITTI to state-of-the-art models trained with 100% human-annotated frames.

To address the limitations of existing pruning methods in practical applications, such as the necessity of training from scratch with sparsity regularization or complex data-driven optimization, we set out from a novel perspective to explore parameter redundancy and accelerate deep CNNs. Precisely, we argue that channels revealing similar feature information have functional overlap and that each such similarity group can be reduced to a few representatives with little impact on the representational power of the model. After deriving an effective metric for evaluating channel similarity via probabilistic modeling, we introduce a similarity-based pruning framework based on hierarchical clustering.In particular, the proposed algorithm can be directly applied to all kinds of pre-trained CNN models for better trade-offs between latency and accuracy. Moreover, rather than relying on a pre-defined target structure, it automatically discovers resource-efficient ones out of the original model under given budgets, which is in the same flavor as NAS. Extensive experiments on benchmark datasets well demonstrate the superior performance of our approach over prior arts. On ImageNet, our pruned ResNet-50 with 30% FLOPs reduced outperforms the original model. We further extend our algorithm to a GAN-based generative model and achieve$2\times $acceleration, showing its remarkable generalization capability and flexibility.

We investigate GAN inversion problems of using pre-trained GANs to reconstruct real images. Recent methods for such problems typically employ a VGG perceptual loss to measure the difference between images. While the perceptual loss has achieved remarkable success in various computer vision tasks, it may cause unpleasant artifacts and is sensitive to changes in input scale. This paper delivers an important message that algorithm details are crucial for achieving satisfying performance. In particular, we propose two important but undervalued design principles: (i) not down-sampling the input of the perceptual loss to avoid high-frequency artifacts; and (ii) calculating the perceptual loss using convolutional features which are robust to scale. Integrating these designs derives the proposed framework, HRInversion, that achieves superior performance in reconstructing image details. We validate the effectiveness of HRInversion on a cross-domain image synthesis task and propose a post-processing approach named local style optimization (LSO) to synthesize clean and controllable stylized images. For the evaluation of the cross-domain images, we introduce a metric named ID retrieval which captures the similarity of face identities of stylized images to content images. We also test HRInversion on non-square images. Equipped with implicit neural representation, HRInversion applies to ultra-high resolution images with more than 10 million pixels. Furthermore, we show applications of style transfer and 3D-aware GAN inversion, paving the way for extending the application range of HRInversion.

Neural network quantization has shown to be an effective way for network compression and acceleration. However, existing binary or ternary quantization methods suffer from two major issues. First, low bit-width input/activation quantization easily results in severe prediction accuracy degradation. Second, network training and quantization are always treated as two non-related tasks, leading to accumulated parameter training error and quantization error. In this work, we introduce a novel scheme, namedResidual Quantization, to train a neural network with both weights and inputs constrained to low bit-width, e.g., binary or ternary values. On one hand, by recursively performing residual quantization, the resulting binary/ternary network is guaranteed to approximate the full-precision network with much smaller errors. On the other hand, we mathematically re-formulate the network training scheme in anEM-likemanner, which iteratively performs network quantization and parameter optimization. Duringexpectation, the low bit-width network is encouraged to approximate the full-precision network. Duringmaximization, the low bit-width network is further tuned to gain better representation capability. Extensive experiments well demonstrate that the proposed quantization scheme outperforms previous low bit-width methods and achieves much closer performance to the full-precision counterpart.

Dynamic convolution has achieved significant gain in performance and computational complexity, thanks to its powerful representation capability given limited filter number/layers. However, SOTA dynamic convolution operators are sensitive to input noises (e.g., Gaussian noise, shot noise, e.t.c.) and lack sufficient spatial contextual information in filter generation. To alleviate this inherent weakness, we propose a lightweight and heterogeneous-structure (i.e., static and dynamic) operator, named Bi-volution. On the one hand, Bi-volution is designed as a dual-branch structure to fully leverage complementary properties of static/dynamic convolution, which endows Bi-volution more robust properties and higher performance. On the other hand, the Spatial Augmented Kernel Generation module is proposed to improve the dynamic convolution, realizing the learning of spatial context information with negligible additional computational complexity. Extensive experiments illustrate that the ResNet-50 equipped with Bi-volution achieves a highly competitive boost in performance (+2.8% top-1 accuracy on ImageNet classification, +2.4% box AP and +2.2% mask AP on COCO detection and instance segmentation) while maintaining extremely low FLOPs (i.e., [email protected] GFLOPs). Furthermore, our Bi-volution shows better robustness than dynamic convolution against various noise and input corruptions. Our code is available at https://github.com/neuralchen/Bivolution.

In this paper, we give a new definition for sample complexity, and further develop a theoretical analysis to bridge the gap between sample complexity and model capacity. In contrast to previous works which study on some toy samples, we conduct our analysis on more general data space, and build a qualitative relationship from sample complexity to model capacity required to achieve comparable performance. Besides, we introduce a simple indicator to evaluate the sample complexity based on continuous mapping. Moreover, we further analysis the relationship between sample complexity and data distribution, which paves the way to understand the present representation learning. Extensive experiments on several datasets well demonstrate the effectiveness of our evaluation method.

Person search has recently emerged as a challenging task that jointly addresses pedestrian detection and person re-identification. Existing approaches follow a fully supervised setting where both bounding box and identity annotations are available. However, annotating identities is labor-intensive, limiting the practicability and scalability of current frameworks. This paper inventively considers weakly supervised person search with only bounding box annotations. We propose to address this novel task by investigating three levels of context clues (i.e., detection, memory and scene) in unconstrained natural images. The first two are employed to promote local and global discriminative capabilities, while the latter enhances clustering accuracy. Despite its simple design, our CGPS boosts the baseline model by 8.8% in mAP on CUHK-SYSU. Surprisingly, it even achieves comparable performance with several supervised person search models. Our code is available at https://github. com/ljpadam/CGPS.

We address cross-species 3D face morphing (i.e., 3D face morphing from human to animal), a novel problem with promising applications in social media and movie industry. It remains challenging how to preserve target structural information and source fine-grained facial details simultaneously. To this end, we propose an Alignment-aware 3D Face Morphing (AFM) framework, which builds semantic-adaptive correspondence between source and target faces across species, via an alignment-aware controller mesh (Explicit Controller, EC) with explicit source/target mesh binding. Based on EC, we introduce Controller-Based Mapping (CBM), which builds semantic consistency between source and target faces according to the semantic importance of different face regions. Additionally, an inference-stage coarse-to-fine strategy is exploited to produce fine-grained meshes with rich facial details from rough meshes. Extensive experimental results in multiple people and animals demonstrate that our method produces high-quality deformation results.

Hierarchical semantic structures naturally exist in an image dataset, in which several semantically relevant image clusters can be further integrated into a larger cluster with coarser-grained semantics. Capturing such structures with image representations can greatly benefit the semantic understanding on various downstream tasks. Existing contrastive representation learning methods lack such an important model capability. In addition, the negative pairs used in these methods are not guaranteed to be semantically distinct, which could further hamper the structural correctness of learned image representations. To tackle these limitations, we propose a novel contrastive learning framework called Hierarchical Contrastive Selective Coding (HCSC). In this framework, a set of hierarchical prototypes are constructed and also dynamically updated to represent the hierarchical semantic structures underlying the data in the latent space. To make image representations better fit such semantic structures, we employ and further improve conventional instance-wise and prototypical contrastive learning via an elaborate pair selection scheme. This scheme seeks to select more diverse positive pairs with similar semantics and more precise negative pairs with truly distinct semantics. On extensive downstream tasks, we verify the state-of-the-art performance of HCSC and also the effectiveness of major model components. We are continually building a comprehensive model zoo (see supplementary material). Our source code and model weights are available at https://github.com/gyfastas/HCSC.

Appearance-based Gaze Estimation leverages deep neural networks to regress the gaze direction from monocular images and achieve impressive performance. However, its success depends on expensive and cumbersome annotation capture. When lacking precise annotation, the large domain gap hinders the performance of trained models on new domains. In this paper, we propose a novel gaze adaptation approach, namely Contrastive Regression Gaze Adaptation (CRGA), for generalizing gaze estimation on the target domain in an unsupervised manner. CRGA leverages the Contrastive Domain Generalization (CDG) module to learn the stable representation from the source domain and leverages the Contrastive Self-training Adaptation (CSA) module to learn from the pseudo labels on the target domain. The core of both CDG and CSA is the Contrastive Regression (CR) loss, a novel contrastive loss for regression by pulling features with closer gaze directions closer together while pushing features with farther gaze directions farther apart. Experimentally, we choose ETH-XGAZE and Gaze-360 as the source domain and test the domain generalization and adaptation performance on MPIIGAZE, RT-GENE, Gaze-Capture, EyeDiap respectively. The results demonstrate that our CRGA achieves remarkable performance improvement compared with the baseline models and also outperforms the state-of-the-art domain adaptation approaches on gaze adaptation tasks.

Deep implicit shape models have become popular in the computer vision community at large but less so for biomed-ical applications. This is in part because large training databases do not exist and in part because biomedical an-notations are often noisy. In this paper, we show that by introducing templates within the deep learning pipeline we can overcome these problems. The proposed framework, named ImplicitAtlas, represents a shape as a deformation field from a learned template field, where multiple templates could be integrated to improve the shape representation ca-pacity at negligible computational cost. Extensive experi-ments on three medical shape datasets prove the superiority over current implicit representation methods.

Human motion synthesis is a long-standing problem with various applications in digital twins and the Metaverse. However, modern deep learning based motion synthesis approaches barely consider the physical plausibility of synthesized motions and consequently they usually produce unrealistic human motions. In order to solve this problem, we propose a system "Skeleton2Humanoid" which performs physics-oriented motion correction at test time by regularizing synthesized skeleton motions in a physics simulator. Concretely, our system consists of three sequential stages: (I) test time motion synthesis network adaptation, (II) skeleton to humanoid matching and (III) motion imitation based on reinforcement learning (RL). Stage I introduces a test time adaptation strategy, which improves the physical plausibility of synthesized human skeleton motions by optimizing skeleton joint locations. Stage II performs an analytical inverse kinematics strategy, which converts the optimized human skeleton motions to humanoid robot motions in a physics simulator, then the converted humanoid robot motions can be served as reference motions for the RL policy to imitate. Stage III introduces a curriculum residual force control policy, which drives the humanoid robot to mimic complex converted reference motions in accordance with the physical law. We verify our system on a typical human motion synthesis task, motion-in-betweening. Experiments on the challenging LaFAN1 dataset show our system can outperform prior methods significantly in terms of both physical plausibility and accuracy. Code will be released for research purposes at: https://github.com/michaelliyunhao/Skeleton2Humanoid.

AI-for-science approaches have been applied to solve scientific problems (e.g., nuclear fusion, ecology, genomics, meteorology) and have achieved highly promising results. Spatial precipitation downscaling is one of the most important meteorological problem and urgently requires the participation of AI. However, the lack of a well-organized and annotated large-scale dataset hinders the training and verification of more effective and advancing deep-learning models for precipitation downscaling. To alleviate these obstacles, we present the first large-scale spatial precipitation downscaling dataset named RainNet, which contains more than 62,400 pairs of high-quality low/high-resolution precipitation maps for over 17 years, ready to help the evolution of deep learning models in precipitation downscaling. Specifically, the precipitation maps carefully collected in RainNet cover various meteorological phenomena (e.g., hurricane, squall), which is of great help to improve the model generalization ability. In addition, the map pairs in RainNet are organized in the form of image sequences (720 maps per month or 1 map/hour), showing complex physical properties, e.g., temporal misalignment, temporal sparse, and fluid properties. Furthermore, two deep-learning-oriented metrics are specifically introduced to evaluate or verify the comprehensive performance of the trained model (e.g., prediction maps reconstruction accuracy). To illustrate the applications of RainNet, 14 state-of-the-art models, including deep models and traditional approaches, are evaluated. To fully explore potential downscaling solutions, we propose an implicit physical estimation benchmark framework to learn the above characteristics. Extensive experiments demonstrate the value of RainNet in training and evaluating downscaling models. Our dataset is available at https://neuralchen.github.io/RainNet/.

Learning to generate continuous linguistic descriptions for multi-subject interactive videos in great details has particular applications in team sports auto-narrative. In contrast to traditional video caption, this task is more challenging as it requires simultaneous modeling of fine-grained individual actions, uncovering of spatio-temporal dependency structures of frequent group interactions, and then accurate mapping of these complex interaction details into long and detailed commentary. To explicitly address these challenges, we propose a novel framework Graph-based Learning for Multi-Granularity Interaction Representation (GLMGIR) for fine-grained team sports auto-narrative task. A multi-granular interaction modeling module is proposed to extract among-subjects' interactive actions in a progressive way for encoding both intra- and inter-team interactions. Based on the above multi-granular representations, a multi-granular attention module is developed to consider action/event descriptions of multiple spatio-temporal resolutions. Both modules are integrated seamlessly and work in a collaborative way to generate the final narrative. In the meantime, to facilitate reproducible research, we collect a new video dataset from YouTube.com called Sports Video Narrative dataset (SVN). It is a novel direction as it contains 6K team sports videos (i.e., NBA basketball games) with 10K ground-truth narratives(e.g., sentences). Furthermore, as previous metrics such as METEOR (i.e., used in coarse-grained video caption task) DO NOT cope with fine-grained sports narrative task well, we hence develop a novel evaluation metric named Fine-grained Captioning Evaluation (FCE), which measures how accurate the generated linguistic description reflects fine-grained action details as well as the overall spatio-temporal interactional structure. Extensive experiments on our SVN dataset have demonstrated the effectiveness of the proposed framework for fine-grained team sports video auto-narrative.

From the perspective of probability, we propose a new method for black-box adversarial attack via black-box variational inference (BBVI), where the knowledge of victim model is unavailable. Instead of obtaining a single point, the proposed method focuses on approximating the probability distribution of adversarial examples. Thus, infinite adversarial examples can be drawn from the inferred distribution. Although the Monte Carlo estimator in BBVI is unbiased, its variance brings unstable gradient estimation, which leads to poor attack performance and low query efficiency. To reduce variance, we improve the BBVI with importance sampling which guided by a surrogate model to obtain a better estimator of gradient, which enhances both success rate and query efficiency. Extensive experiments on ImageNet dataset well demonstrate the outperformance of the proposed method compared with prior arts.

Conditional generative adversarial networks (cGANs) are designed to generate images based on the provided conditions,e.g., class-level distributions, semantic label maps,etc. Existing methods have used the same generator architecture for all classes. This paper presents an idea that adopts neural architecture search (NAS) to find a class-aware architecture for each class. The search space contains regular and class-modulated convolutions, where the latter is designed to introduce class-specific information while avoiding the reduction of training data for each class generator. The search algorithm follows a weight-sharing pipeline with mixed-architecture optimization so that the search cost does not grow with the number of classes. To learn the sampling policy, a Markov decision process is embedded into the search algorithm, and a moving average is applied for better stability. Class-aware generators show advantages over class-agnostic architectures experimentally. Moreover, we discover two intriguing phenomena that are inspirational to craft cGANs by hand.

We propose a novel image-to-pencil translation method that could not only generate high-quality pencil sketches but also offer the drawing process. Existing pencil sketch algorithms are based on texture rendering rather than the direct imitation of strokes, making them unable to show the drawing process but only a final result. To address this challenge, we first establish a pencil stroke imitation mechanism. Next, we develop a framework with three branches to guide stroke drawing: the first branch guides the direction of the strokes, the second branch determines the shade of the strokes, and the third branch enhances the details further. Under this framework's guidance, we can produce a pencil sketch by drawing one stroke every time. Our method is fully interpretable. Comparison with existing pencil drawing algorithms shows that our method is superior to others in terms of texture quality, style, and user evaluation. Our code and supplementary material are now available at: https://github.com/TZYSJTU/Sketch-Generation-withDrawing-Process-Guided-by-Vector-Flow-and-Grayscale

This paper considers a new problem of adapting a pretrained model of human mesh reconstruction to out-of-domain streaming videos. However, most previous methods based on the parametric SMPL model [36] underperform in new domains with unexpected, domain-specific attributes, such as camera parameters, lengths of bones, backgrounds, and occlusions. Our general idea is to dynamically fine-tune the source model on test video streams with additional temporal constraints, such that it can mitigate the domain gaps without over-fitting the 2D information of individual test frames. A subsequent challenge is how to avoid conflicts between the 2D and temporal constraints. We propose to tackle this problem using a new training algorithm named Bilevel Online Adaptation (BOA), which divides the optimization process of overall multi-objective into two steps of weight probe and weight update in a training iteration. We demonstrate that BOA leads to state-of-the-art results on two human mesh reconstruction benchmarks1.

Unsupervised learning methods have recently shown their competitiveness against supervised training. Typically, these methods use a single objective to train the en-tire network. But one distinct advantage of unsupervised over supervised learning is that the former possesses more variety and freedom in designing the objective. In this work, we explore new dimensions of unsupervised learning by proposing the Progressive Stage-wise Learning (PSL) framework. For a given unsupervised task, we design multi-level tasks and define different learning stages for the deep network. Early learning stages are forced to focus on low-level tasks while late stages are guided to extract deeper information through harder tasks. We discover that by progressive stage-wise learning, unsupervised feature representation can be effectively enhanced. Our extensive experiments show that PSL consistently improves results for the leading unsupervised learning methods.

In this work, we propose a Cross-view Contrastive Learning framework for unsupervised 3D skeleton-based action Representation (CrosSCLR), by leveraging multi-view complementary supervision signal. CrosSCLR consists of both single-view contrastive learning (Skeleton-CLR) and cross-view consistent knowledge mining (CVC-KM) modules, integrated in a collaborative learning manner. It is noted that CVC-KM works in such a way that high-confidence positive/negative samples and their distributions are exchanged among views according to their embedding similarity, ensuring cross-view consistency in terms of contrastive context, i.e., similar distributions. Extensive experiments show that CrosSCLR achieves remarkable action recognition results on NTU-60 and NTU-120 datasets under unsupervised settings, with observed higher-quality action representations. Our code is available at https://github.com/LinguoLi/CrosSCLR.

We develop a novel self-supervised learning method named Shape Self-Correction for point cloud analysis. Our method is motivated by the principle that a good shape representation should be able to find distorted parts of a shape and correct them. To learn strong shape representations in an unsupervised manner, we first design a shape-disorganizing module to destroy certain local shape parts of an object. Then the destroyed shape and the normal shape are sent into a point cloud network to get representations, which are employed to segment points that belong to distorted parts and further reconstruct them to restore the shape to normal. To perform better in these two associated pretext tasks, the network is constrained to capture useful shape features from the object, which indicates that the point cloud network encodes rich geometric and contextual information. The learned feature extractor transfers well to downstream classification and segmentation tasks. Experimental results on ModelNet, ScanNet and ShapeNetPart demonstrate that our method achieves state-of-the-art performance among unsupervised methods. Our framework can be applied to a wide range of deep learning networks for point cloud analysis and we show experimentally that pre-training with our framework significantly boosts the performance of supervised models.

Pixel-to-mesh has wide applications, especially in virtual or augmented reality, animation and game industry. However, existing mesh reconstruction models perform unsatisfactorily in local geometry details due to ignoring mesh topology information during learning. Besides, most methods are constrained by the initial template, which cannot reconstruct meshes of various genus. In this work, we propose a geometric granularity-aware pixel-to-mesh framework with a fidelity-selection-and-guarantee strategy, which explicitly addresses both challenges. First, a geometry structure extractor is proposed for detecting local high structured parts and capturing local spatial feature. Second, we apply it to facilitate pixel-to-mesh mapping and resolve coarse details problem caused by the neglect of structural information in previous practices. Finally, a mesh edit module is proposed to encourage non-zero genus topology to emergence by fine-grained topology modification and a patching algorithm is introduced to repair the non-closed boundaries. Extensive experimental results, both quantitatively and visually have demonstrated the high reconstruction fidelity achieved by the proposed framework.

Autonomous highlight detection is crucial for enhancing the efficiency of video browsing on social media platforms. To attain this goal in a data-driven way, one may often face the situation where highlight annotations are not available on the target video category used in practice, while the supervision on another video category (named as source video category) is achievable. In such a situation, one can derive an effective highlight detector on target video category by transferring the highlight knowledge acquired from source video category to the target one. We call this problem cross-category video highlight detection, which has been rarely studied in previous works. For tackling such practical problem, we propose a Dual-Learner-based Video Highlight Detection (DL-VHD) framework. Under this framework, we first design a Set-based Learning module (SL-module) to improve the conventional pair-based learning by assessing the highlight extent of a video segment under a broader context. Based on such learning manner, we introduce two different learners to acquire the basic distinction of target category videos and the characteristics of highlight moments on source video category, respectively. These two types of highlight knowledge are further consolidated via knowledge distillation. Extensive experiments on three benchmark datasets demonstrate the superiority of the proposed SL-module, and the DL-VHD method outperforms five typical Unsupervised Domain Adaptation (UDA) algorithms on various cross-category highlight detection tasks. Our code is available at https://github.com/ChrisAllenMing/Cross_Category_Video_Highlight.

In this work, we study the ambiguity problem in the task of unsupervised 3D human pose estimation from 2D counterpart. On one hand, without explicit annotation, the scale of 3D pose is difficult to be accurately captured (scale ambiguity). On the other hand, one 2D pose might correspond to multiple 3D gestures, where the lifting procedure is inherently ambiguous (pose ambiguity). Previous methods generally use temporal constraints (e.g., constant bone length and motion smoothness) to alleviate the above issues. However, these methods commonly enforce the outputs to fulfill multiple training objectives simultaneously, which often lead to sub-optimal results. In contrast to the majority of previous works, we propose to split the whole problem into two sub-tasks, i.e., optimizing 2D input poses via a scale estimation module and then mapping optimized 2D pose to 3D counterpart via a pose lifting module. Furthermore, two temporal constraints are proposed to alleviate the scale and pose ambiguity respectively. These two modules are optimized via a iterative training scheme with corresponding temporal constraints, which effectively reduce the learning difficulty and lead to better performance. Results on the Human3.6M dataset demonstrate that our approach improves upon the prior art by 23.1% and also outperforms several weakly supervised approaches that rely on 3D annotations. Our project is available at https://sites.google.com/view/ambiguity-aware-hpe.

In this paper, we decouple unsupervised human mesh recovery into the well-studied problems of unsupervised 3D pose estimation, and human mesh recovery from estimated 3D skeletons, focusing on the latter task. The challenges of the latter task are two folds: (1) pose failure (i.e., pose mismatching – different skeleton definitions in dataset and SMPL , and pose ambiguity – endpoints have arbitrary joint angle configurations for the same 3D joint coordinates). (2) shape ambiguity (i.e., the lack of shape constraints on body configuration). To address these issues, we propose Skeleton2Mesh, a novel lightweight framework that recovers human mesh from a single image. Our Skeleton2Mesh contains three modules, i.e., Differentiable Inverse Kinematics (DIK), Pose Refinement (PR) and Shape Refinement (SR) modules. DIK is designed to transfer 3D rotation from estimated 3D skeletons, which relies on a minimal set of kinematics prior knowledge. Then PR and SR modules are utilized to tackle the pose ambiguity and shape ambiguity respectively. All three modules can be incorporated into Skeleton2Mesh seamlessly via an end-to-end manner. Furthermore, we utilize an adaptive joint regressor to alleviate the effects of skeletal topology from different datasets. Results on the Human3.6M dataset for human mesh recovery demonstrate that our method improves upon the previous unsupervised methods by 32.6% under the same setting. Qualitative results on in-the-wild datasets exhibit that the recovered 3D meshes are natural, realistic. Our project is available at https://sites.google.com/view/skeleton2mesh.

The conditional generative adversarial network (cGAN) is a powerful tool of generating high-quality images, but existing approaches mostly suffer unsatisfying performance or the risk of mode collapse. This paper presents Omni-GAN, a variant of cGAN that reveals the devil in designing a proper discriminator for training the model. The key is to ensure that the discriminator receives strong supervision to perceive the concepts and moderate regularization to avoid collapse. Omni-GAN is easily implemented and freely integrated with off-the-shelf encoding methods (e.g., implicit neural representation, INR). Experiments validate the superior performance of Omni-GAN and Omni-INR-GAN in a wide range of image generation and restoration tasks. In particular, Omni-INR-GAN sets new records on the ImageNet dataset with impressive Inception scores of 262.85 and 343.22 for the image sizes of 128 and 256, respectively, surpassing the previous records by 100+ points. Moreover, leveraging the generator prior, Omni-INR-GAN can extrapolate low-resolution images to arbitrary resolution, even up to ×60+ higher resolution. Code is available1.

This paper studies unsupervised/self-supervised whole-graph representation learning, which is critical in many tasks such as molecule properties prediction in drug and material discovery. Existing methods mainly focus on preserving the local similarity structure between different graph instances but fail to discover the global semantic structure of the entire data set. In this paper, we propose a unified framework called Local-instance and Global-semantic Learning (GraphLoG) for self-supervised whole-graph representation learning. Specifically, besides preserving the local similarities, GraphLoG introduces the hierarchical prototypes to capture the global semantic clusters. An efficient online expectation-maximization (EM) algorithm is further developed for learning the model. We evaluate GraphLoG by pre-training it on massive unlabeled graphs followed by fine-tuning on downstream tasks. Extensive experiments on both chemical and biological benchmark data sets demonstrate the effectiveness of the proposed approach.

Recent advances in image inpainting have shown impressive results for generating plausible visual details on rather simple backgrounds. However, for complex scenes, it is still challenging to restore reasonable contents as the contextual information within the missing regions tends to be ambiguous. To tackle this problem, we introduce pretext tasks that are semantically meaningful to estimating the missing contents. In particular, we perform knowledge distillation on pretext models and adapt the features to image inpainting. The learned semantic priors ought to be partially invariant between the high-level pretext task and low-level image inpainting, which not only help to understand the global context but also provide structural guidance for the restoration of local textures. Based on the semantic priors, we further propose a context-aware image inpainting model, which adaptively integrates global semantics and local features in a unified image generator. The semantic learner and the image generator are trained in an end-to-end manner. We name the model SPL to highlight its ability to learn and leverage semantic priors. It achieves the state of the art on Places2, CelebA, and Paris StreetView datasets

An in-depth scene understanding usually requires recognizing all the objects and their relations in an image, encoded as a scene graph. Most existing approaches for scene graph generation first independently recognize each object and then predict their relations independently. Though these approaches are very efficient, they ignore the dependency between different objects as well as between their relations. In this paper, we propose a principled approach to jointly predict the entire scene graph by fully capturing the dependency between different objects and between their relations. Specifically, we establish a unified conditional random field (CRF) to model the joint distribution of all the objects and their relations in a scene graph. We carefully design the potential functions to enable relational reasoning among different objects according to knowledge graph embedding methods. We further propose an efficient and effective algorithm for inference based on mean-field variational inference, in which we first provide a warm initialization by independently predicting the objects and their relations according to the current model, followed by a few iterations of relational reasoning. Experimental results on both the relationship retrieval and zero-shot relationship retrieval tasks prove the efficiency and efficacy of our proposed approach.

In this paper, we propose an adaptive region proposal scheme with feature channel regularization to facilitate robust object tracking. We consider tracking as a linear regression problem and an ensemble of correlation filters is trained on-line to distinguish the foreground target from the background. Further, we integrate adaptively learned region proposals into an enhanced two-stream tracking framework based on correlation filters. For the tracking stream, we learn two-stage cascade correlation filters on deep convolutional features to ensure competitive tracking performance. For the detection stream, we employ adaptive region proposals, which are effective in recovering target objects from tracking failures caused by heavy occlusion or out-of-view movement. In contrast to traditional tracking-by-detection methods using random samples or sliding windows, we perform target re-detection over adaptively learned region proposals. Since region proposals naturally take the objectness information into account, we show that the proposed adaptive region proposals can handle the challenging scale estimation problem as well. In addition, we observe the channel redundancy and noisy of feature representation, especially for the convolutional features. Thus, we apply a channel regularization to the correlation filter learning. Extensive experimental validations on OTB, VOT and UAV-123 datasets demonstrate that the proposed method performs favorably against state-of-the-art tracking algorithms.

There have been considerable debates over 2D and 3D representation learning on 3D medical images. 2D approaches could benefit from large-scale 2D pretraining, whereas they are generally weak in capturing large 3D contexts. 3D approaches are natively strong in 3D contexts, however few publicly available 3D medical dataset is large and diverse enough for universal 3D pretraining. Even for hybrid (2D + 3D) approaches, the intrinsic disadvantages within the 2D/3D parts still exist. In this study, we bridge the gap between 2D and 3D convolutions by reinventing the 2D convolutions. We propose ACS (axial-coronal-sagittal) convolutions to perform natively 3D representation learning, while utilizing the pretrained weights on 2D datasets. In ACS convolutions, 2D convolution kernels are split by channel into three parts, and convoluted separately on the three views (axial, coronal and sagittal) of 3D representations. Theoretically, ANY 2D CNN (ResNet, DenseNet, or DeepLab) is able to be converted into a 3D ACS CNN, with pretrained weight of a same parameter size. Extensive experiments validate the consistent superiority of the pretrained ACS CNNs, over the 2D/3D CNN counterparts with/without pretraining. Even without pretraining, the ACS convolution can be used as a plug-and-play replacement of standard 3D convolution, with smaller model size and less computation.

Recent works on domain adaptation reveal the effectiveness of adversarial learning on filling the discrepancy between source and target domains. However, two common limitations exist in current adversarial-learning-based methods. First, samples from two domains alone are not sufficient to ensure domain-invariance at most part of latent space. Second, the domain discriminator involved in these methods can only judge real or fake with the guidance of hard label, while it is more reasonable to use soft scores to evaluate the generated images or features, i.e., to fully utilize the inter-domain information. In this paper, we present adversarial domain adaptation with domain mixup (DM-ADA), which guarantees domain-invariance in a more continuous latent space and guides the domain discriminator in judging samples' difference relative to source and target domains. Domain mixup is jointly conducted on pixel and feature level to improve the robustness of models. Extensive experiments prove that the proposed approach can achieve superior performance on tasks with various degrees of domain shift and data complexity.

Applying the knowledge of an object detector trained on a specific domain directly onto a new domain is risky, as the gap between two domains can severely degrade model's performance. Furthermore, since different instances commonly embody distinct modal information in object detection scenario, the feature alignment of source and target domain is hard to be realized. To mitigate these problems, we propose a Graph-induced Prototype Alignment (GPA) framework to seek for category-level domain alignment via elaborate prototype representations. In the nutshell, more precise instance-level features are obtained through graph-based information propagation among region proposals, and, on such basis, the prototype representation of each class is derived for category-level domain alignment. In addition, in order to alleviate the negative effect of class-imbalance on domain adaptation, we design a Class-reweighted Contrastive Loss to harmonize the adaptation training process. Combining with Faster R-CNN, the proposed framework conducts feature alignment in a two-stage manner. Comprehensive results on various cross-domain detection tasks demonstrate that our approach outperforms existing methods with a remarkable margin. Our code is available at https://github.com/ChrisAllenMing/GPA-detection.

For monocular 3D pose estimation conditioned on 2D detection, noisy/unreliable input is a key obstacle in this task. Simple structure constraints attempting to tackle this problem, e.g., symmetry loss and joint angle limit, could only provide marginal improvements and are commonly treated as auxiliary losses in previous researches. Thus it still remains challenging about how to effectively utilize the power of human prior knowledge for this task. In this paper, we propose to address above issue in a systematic view. Firstly, we show that optimizing the kinematics structure of noisy 2D inputs is critical to obtain accurate 3D estimations. Secondly, based on corrected 2D joints, we further explicitly decompose articulated motion with human topology, which leads to more compact 3D static structure easier for estimation. Finally, temporal refinement emphasizing the validity of 3D dynamic structure is naturally developed to pursue more accurate result. Above three steps are seamlessly integrated into deep neural models, which form a deep kinematics analysis pipeline concurrently considering the static/dynamic structure of 2D inputs and 3D outputs. Extensive experiments show that proposed framework achieves state-of-the-art performance on two widely used 3D human action datasets. Meanwhile, targeted ablation study shows that each former step is critical for the latter one to obtain promising results.

In video prediction tasks, one major challenge is to capture the multi-modal nature of future contents and dynamics. In this work, we propose a simple yet effective framework that can efficiently predict plausible future states, where the key insight is that the potential distribution of a sequence could be approximated with analogous ones in a repertoire of training pool, namely, expert examples. By further incorporating a novel optimization scheme into the training procedure, plausible predictions can be sampled efficiently from distribution constructed from the retrieved examples. Meanwhile, our method could be seamlessly integrated with existing stochastic predictive models; significant enhancement is observed with comprehensive experiments in both quantitative and qualitative aspects. We also demonstrate the generalization ability to predict the motion of unseen class, i.e., without access to corresponding data during training phase. Project Page: \hyperlink{https://sites.google.com/view/vpeg-supp/home.}{https://sites.google.com/view/vpeg-supp/home.}

This paper proposes a simple yet effective strategy for improving deep supervised learning, named Stochastic Label Refinery (SLR), by refining training labels to more informative labels. When training a neural network, target distributions (or ground-truth) are typically “hard”, which means the target label of each category consists of only 0 and 1. However, the fixed “hard” target distributions do not capture association between categories or that between objects. In this study, instead of using the hard target distributions, we iteratively generate “soft” target label distributions for training the neural networks, which leads to better performances. The soft target distributions are obtained via an Expectation-Maximization (EM) iteration, where the “true” target distributions and the learned models are regarded as hidden variables. In E step, the models are optimized to approximate the target distributions on stochastic splits of training data; In M step, the target distributions are updated with predicted pseudo-label on leave-out splits. Extensive experiments on classification and ordinal regression tasks, empirically prove that the refined target distribution consistently leads to considerable performance improvements even applied on competitive baselines. Notably, in DeepDR 2020 Diabetic Retinopathy Grading (DeepDRiD) challenge, our method improves the quadratic weighted kappa on official validation set from 0.8247 to 0.8348 and achieves a state-of-the-art score on online test set. The proposed SLR technique is easy to implement and practically applicable.

We propose an efficient framework, called Simple Swap (SimSwap), aiming for generalized and high fidelity face swapping. In contrast to previous approaches that either lack the ability to generalize to arbitrary identity or fail to preserve attributes like facial expression and gaze direction, our framework is capable of transferring the identity of an arbitrary source face into an arbitrary target face while preserving the attributes of the target face. We overcome the above defects in the following two ways. First, we present the ID Injection Module (IIM) which transfers the identity information of the source face into the target face at feature level. By using this module, we extend the architecture of an identity-specific face swapping algorithm to a framework for arbitrary face swapping. Second, we propose the Weak Feature Matching Loss which efficiently helps our framework to preserve the facial attributes in an implicit way. Extensive experiments on wild faces demonstrate that our SimSwap is able to achieve competitive identity performance while preserving attributes better than previous state-of-the-art methods.

Though significant progress has been made in artistic style transfer, semantic information is usually difficult to be preserved in a fine-grained locally consistent manner by most existing methods, especially when multiple artists styles are required to transfer within one single model. To circumvent this issue, we propose a Stroke Control Multi-Artist Style Transfer framework. On the one hand, we design an Anisotropic Stroke Module (ASM) which realizes the dynamic adjustment of style-stroke between the non-trivial and the trivial regions. ASM endows the network with the ability of adaptive semantic-consistency among various styles. On the other hand, we present an novel Multi-Scale Projection Discriminator to realize the texture-level conditional generation. In contrast to the single-scale conditional discriminator, our discriminator is able to capture multi-scale texture clue to effectively distinguish a wide range of artistic styles. Extensive experimental results well demonstrate the feasibility and effectiveness of our approach. Our framework can transform a photograph into different artistic style oil painting via only ONE single model. Furthermore, the results are with distinctive artistic style and retain the anisotropic semantic information.

This paper addresses the challenging black-box adversarial attack problem, where only classification confidence of a victim model is available. Inspired by consistency of visual saliency between different vision models, a surrogate model is expected to improve the attack performance via transferability. By combining transferability-based and query-based black-box attack, we propose a surprisingly simple baseline approach (named SimBA++) using the surrogate model, which significantly outperforms several state-of-the-art methods. Moreover, to efficiently utilize the query feedback, we update the surrogate model in a novel learning scheme, named High-Order Gradient Approximation (HOGA). By constructing a high-order gradient computation graph, we update the surrogate model to approximate the victim model in both forward and backward pass. The SimBA++ and HOGA result in Learnable Black-Box Attack (LeBA), which surpasses previous state of the art by considerable margins: the proposed LeBA significantly reduces queries, while keeping higher attack success rates close to 100% in extensive ImageNet experiments, including attacking vision benchmarks and defensive models. Code is open source at https://github.com/TrustworthyDL/LeBA.

Scale variation of pedestrians in a crowd image presents a significant challenge for vision-based people counting systems. Such variations are mainly caused by perspective-related distortions due to the camera pose relative to the ground plane. Following the density-based counting paradigm, we postulate that generating density values adaptive to object scales plays a critical role in the accuracy of the final counting results. Motivated by this, we distill the underlying information from depth cues to obtain scale-aware representations that can respond to object scales considering the fact that the scale is inversely proportional to the object depth. Specifically, we propose a depth embedding module as add-ons into existing networks. This module exploits essential depth cues to spatially re-calibrate the magnitude of the original features. In this way, the objects, although in the same class, will attain distinct representations according to their scales, which directly benefits the estimation of scale-aware density values. We conduct a comprehensive analysis of the effects of the depth embedding module and validate that exploiting depth cues to perceive object scale variations in convolutional neural networks improves crowd counting performances. Our experiments demonstrate the effectiveness of the proposed approach on four popular benchmark datasets.

Being able to accurately predict the common gaze point of a group of persons is of particular interest to precise marketing and automatic group attention assessment. Group gaze estimation faces challenges including small face/head size and outlier observers. To address these challenges, we proposed a novel framework called Multi-stream Group Gaze Learning and Estimation (MUGGLE). The MUGGLE infrastructure includes two inference streams: 1) a holistic stream which utilizes fused attention map as input to a global deep convolutional structure to explore the global geometric configurations and contexts of interesting persons in the scene; and 2) an aggregative stream which robustly aggregates individual gazes via a recurrent structure (e.g., LSTM) to obtain outlier-tolerant estimation. Both streams are seamlessly integrated via a fusion network. Extensive experiments are performed on a fully annotated group gaze image dataset with 8,000+ images and 100,000+ faces (which is publicly releasable). The results demonstrate the effectiveness of the proposed MUGGLE framework in group gaze estimation.

Hashing has been widely used in large-scale image retrieval based on approximate nearest neighbor search. Most learning-to-hashing methods adopt a two-stage algorithm to generate binary codes. First, original images are mapped into continuous visual features. Then, binary codes are generated by quantization step or separate projection. Nevertheless, these methods are sensitive to quantization operation, i.e., thresholding. To explicitly address this issue, this study proposes a novel feature quantization scheme with a loopy recurrent neural network, called loopy residual hashing, for the purpose of high accuracy in image retrieval. Instead of one-off thresholding-based feature binarization, the proposed approach performs an iterative threshold-then-approximate operation, which calculates the quantization residual after each thresholding step and then imitates another round of binarization to further approximate the coding residual. The resulting sequences of binary codes possess higher representation accuracy and extensive experiments on image retrieval demonstrate its superior discriminative capability over the prior art. In the meantime, theoretical approximation error analysis is given.

We present a practical and effective method for human action transfer. Given a sequence of source action and limited target information, we aim to transfer motion from source to target. Although recent works based on GAN or VAE achieved impressive results for action transfer in 2D, there still exists a lot of problems which cannot be avoided, such as distorted and discontinuous human body shape, blurry cloth texture and so on. In this paper, we try to solve these problems in a novel 3D viewpoint. On the one hand, we design a skeleton-to-3D-mesh generator to generate the 3D model, which achieves huge improvement on appearance reconstruction. Furthermore, we add a temporal connection to improve the smoothness of the model. On the other hand, instead of directly utilizing the image in RGB space, we transform the target appearance information into UV space for further pose transformation. Specially, unlike conventional graphics render method directly projects visible pixels to UV space, our transformation is according to pixel’s semantic information. We perform experiments on Human3.6M and HumanEva-I to evaluate the performance of pose generator. Both qualitative and quantitative results show that our method outperforms methods based on generation method in 2D. Additionally, we compare our render method with graphic methods on Human3.6M and People-snapshot. The comparison results show that our render method is more robust and effective.

Quantization has shown stunning efficiency on deep neural network, especially for portable devices with limited resources. Most existing works uncritically extend weight quantization methods to activations. However, we take the view that best performance can be obtained by applying different quantization methods to weights and activations respectively. In this paper, we design a new activation function dubbed CReLU from the quantization perspective and further complement this design with appropriate initialization method and training procedure. Moreover, we develop a specific quantization strategy in which we formulate the forward and backward approximation of weights with binary values and quantize the activations to low bitwdth using linear or logarithmic quantizer. We show, for the first time, our final quantized model with binary weights and ultra low bitwidth activations outperforms the previous best models by large margins on ImageNet as well as achieving nearly a 10.85× theoretical speedup with ResNet-18. Furthermore, ablation experiments and theoretical analysis demonstrate the effectiveness and robustness of CReLU in comparison with other activation functions.

Person re-identification has achieved great progress with deep convolutional neural networks. However, most previous methods focus on learning individual appearance feature embedding, and it is hard for the models to handle difficult situations with different illumination, large pose variance and occlusion. In this work, we take a step further and consider employing context information for person search. For a probe-gallery pair, we first propose a contextual instance expansion module, which employs a relative attention module to search and filter useful context information in the scene. We also build a graph learning framework to effectively employ context pairs to update target similarity. These two modules are built on top of a joint detection and instance feature learning framework, which improves the discriminativeness of the learned features. The proposed framework achieves state-of-the-art performance on two widely used person search datasets.

Geometric deep learning is increasingly important thanks to the popularity of 3D sensors. Inspired by the recent advances in NLP domain, the self-attention transformer is introduced to consume the point clouds. We develop Point Attention Transformers (PATs), using a parameter-efficient Group Shuffle Attention (GSA) to replace the costly Multi-Head Attention. We demonstrate its ability to process size-varying inputs, and prove its permutation equivariance. Besides, prior work uses heuristics dependence on the input data (e.g., Furthest Point Sampling) to hierarchically select subsets of input points. Thereby, we for the first time propose an end-to-end learnable and task-agnostic sampling operation, named Gumbel Subset Sampling (GSS), to select a representative subset of input points. Equipped with Gumbel-Softmax, it produces a "soft" continuous subset in training phase, and a "hard" discrete subset in test phase. By selecting representative subsets in a hierarchical fashion, the networks learn a stronger representation of the input sets with lower computation cost. Experiments on classification and segmentation benchmarks show the effectiveness and efficiency of our methods. Furthermore, we propose a novel application, to process event camera stream as point clouds, and achieve a state-of-the-art performance on DVS128 Gesture Dataset.

We propose a variational Bayesian scheme for pruning convolutional neural networks in channel level. This idea is motivated by the fact that deterministic value based pruning methods are inherently improper and unstable. In a nutshell, variational technique is introduced to estimate distribution of a newly proposed parameter, called channel saliency, based on this, redundant channels can be removed from model via a simple criterion. The advantages are two-fold: 1) Our method conducts channel pruning without desire of re-training stage, thus improving the computation efficiency. 2) Our method is implemented as a stand-alone module, called variational pruning layer, which can be straightforwardly inserted into off-the-shelf deep learning packages, without any special network design. Extensive experimental results well demonstrate the effectiveness of our method: For CIFAR-10, we perform channel removal on different CNN models up to 74\% reduction, which results in significant size reduction and computation saving. For ImageNet, about 40% channels of ResNet-50 are removed without compromising accuracy.

Many previous works on point sets learning achieve excellent performance with hierarchical architecture. Their strategies towards points agglomeration, however, only perform points sampling and grouping in original Euclidean space in a fixed way. These heuristic and task-irrelevant strategies severely limit their ability to adapt to more varied scenarios. To this end, we develop a novel hierarchical point sets learning architecture, with dynamic points agglomeration. By exploiting the relation of points in semantic space, a module based on graph convolution network is designed to learn a soft points cluster agglomeration. We construct a hierarchical architecture that gradually agglomerates points by stacking this learnable and lightweight module. In contrast to fixed points agglomeration strategy, our method can handle more diverse situations robustly and efficiently. Moreover, we propose a parameter sharing scheme for reducing memory usage and computational burden induced by the agglomeration module. Extensive experimental results on several point cloud analytic tasks, including classification and segmentation, well demonstrate the superior performance of our dynamic hierarchical learning framework over current state-of-the-art methods.

We propose a variational Bayesian framework for enhancing few-shot learning performance. This idea is motivated by the fact that single point based metric learning approaches are inherently noise-vulnerable and easy-to-be-biased. In a nutshell, stochastic variational inference is invoked to approximate bias-eliminated class specific sample distributions. In the meantime, a classifier-free prediction is attained by leveraging the distribution statistics on novel samples. Extensive experimental results on several benchmarks well demonstrate the effectiveness of our distribution-driven few-shot learning framework over previous point estimates based methods, in terms of superior classification accuracy and robustness.

A novel single-image rain removal method is proposed based on multi-scale cascading image generation (MSCG). In particular, the proposed method consists of an encoder extracting multi-scale features from images and a decoder generating de-rained images with a cascading mechanism. The encoder ensembles the convolution neural networks using the kernels with different sizes, and integrates their outputs across different scales. The decoder implements a coarse-to-fine image generation framework, adding fine details incrementally to the final de-rained images according to the spatial contextual information on different scales. We test the proposed method on both synthetic and real-world datasets. Experimental results show that the proposed method is robust to the changes of scene, e.g., the viewpoint and the depth, the heaviness of rain, etc., which suppresses the blurring problem of de-rained image and outperforms state-of-the-art methods consistently.

Lack of large available datasets fully annotated is a fundamental bottleneck in pulmonary nodule detection, especially when the corresponding computed tomography(CT) images obtained are device-dependent. We propose a spatio-semantic attentive CycleGAN (SSA-CycleGAN) capable of aligning modalities, as well as distinguishing nodule vs. non-nodule, which in turn achieves effective data augmentation. Specifically, a novel training loss function is established, providing a constraint for semantic preservation and local fidelity of nodule regions. Extensive experimental results on varied datasets demonstrate the proposed framework achieves significant performance gain on pulmonary nodule detection.

Hashing is a widely adopted method based on an approximate nearest neighbor search and is used in large-scale image retrieval tasks. Conventional learning-based hashing algorithms employ end-to-end representation learning, which is a one-off technique. Because of the tradeoff between efficiency and performance, conventional learning-based hashing methods must sacrifice code length to improve performance, which increases their computational complexity. To improve the efficiency of binary codes, motivated by the “nonsalient-to-salient” attention scheme of humans, we propose a recursive hashing mechanism that maps progressively expanded salient regions to a series of binary codes. These salient regions are generated by a conventional saliency model based on bottom-up saliency-driven attention and a semantic-guided saliency model based on top-down task-driven attention. After obtaining a series of salient regions, we perform long-range temporal modeling of salient regions using a graph-based recurrent deep network to obtain more refined representative features. The later output nodes inherit aggregated information from all previous nodes and extract discriminative features from more salient regions. Therefore, this network possesses more significant information and satisfactory scalability. The proposed recursive hashing neural network, optimized by a triplet ranking loss, is end-to-end trainable. Extensive experimental results from several image retrieval benchmarks show the scalability of our method and demonstrate its strong performance compared with state-of-the-art methods.

Video generation is a challenging task due to the extremely high-dimensional distribution of the solution space. Good constraints in the solution domain would thus reduce the difficulty of approximating optimal solutions. In this paper, instead of directly generating high-dimensional video data, we propose using object landmarks as explicit structure constraints to address this issue. Specifically, we propose a two-stage framework for an action-conditioned video generation task. In our framework, the first stage aims to generate landmark sequences according to predefined motion types, and a recurrent model (RNN/LSTM) is adopted for this purpose. The landmark sequence can be regarded as a low-dimensional structure embedding of high-dimensional video data, and generating landmark sequences is much easier than generating videos. The second stage is inspired by a conditional generative adversarial network (CGAN), and we take the generated landmark sequence as a structure condition to learn a landmark-to-image translation network. Such a one-to-one translation framework avoids the difficulty of generating videos and instead transfers the video generation task to image generation, which is resolvable due to the maturity of current GAN-based models. The experimental results demonstrate that our model not only achieves promising results on rigid/nonrigid motion generation tasks but also can be extended to multiobject motion situations.

Action recognition is an important yet challenging task in computer vision. In this paper, we propose a novel deep-based framework for action recognition, which improves the recognition accuracy by: 1) deriving more precise features for representing actions, and 2) reducing the asynchrony between different information streams. We first introduce a coarse-to-fine network which extracts shared deep features at different action class granularities and progressively integrates them to obtain a more accurate feature representation for input actions. We further introduce an asynchronous fusion network. It fuses information from different streams by asynchronously integrating stream-wise features at different time points, hence better leveraging the complementary information in different streams. Experimental results on action recognition benchmarks demonstrate that our approach achieves the state-of-the-art performance.

The goal of video summarization is to distill a raw video into a more compact form without losing much semantic information. However, previous methods mainly consider the diversity and representation interestingness of the obtained summary, and they seldom pay sufficient attention to semantic information of resulting frame set, especially the long temporal range semantics. To explicitly address this issue, we propose a novel technique which is able to extract the most semantically relevant video segments (i.e., valid for a long term temporal duration) and assemble them into an informative summary. To this end, we develop a semantic attended video summarization network (SASUM) which consists of a frame selector and video descriptor to select an appropriate number of video shots by minimizing the distance between the generated description sentence of the summarized video and the human annotated text of the original video. Extensive experiments show that our method achieves a superior performance gain over previous methods on two benchmark datasets.

Person re-identification (ReID) is an important task in the field of intelligent security. A key challenge is how to capture human pose variations, while existing benchmarks (i.e., Market1501, DukeMTMC-reID, CUHK03, etc.) do NOT provide sufficient pose coverage to train a robust ReID system. To address this issue, we propose a pose-transferrable person ReID framework which utilizes pose-transferred sample augmentations (i.e., with ID supervision) to enhance ReID model training. On one hand, novel training samples with rich pose variations are generated via transferring pose instances from MARS dataset, and they are added into the target dataset to facilitate robust training. On the other hand, in addition to the conventional discriminator of GAN (i.e., to distinguish between REAL/FAKE samples), we propose a novel guider sub-network which encourages the generated sample (i.e., with novel pose) towards better satisfying the ReID loss (i.e., cross-entropy ReID loss, triplet ReID loss). In the meantime, an alternative optimization procedure is proposed to train the proposed Generator-Guider-Discriminator network. Experimental results on Market-1501, DukeMTMC-reID and CUHK03 show that our method achieves great performance improvement, and outperforms most state-of-the-art methods without elaborate designing the ReID model.

Crowd counting or density estimation is a challenging task in computer vision due to large scale variations, perspective distortions and serious occlusions, etc. Existing methods generally suffer from two issues: 1) the model averaging effects in multi-scale CNNs induced by the widely adopted ℓ2regression loss; and 2) inconsistent estimation across different scaled inputs. To explicitly address these issues, we propose a novel crowd counting (density estimation) framework called Adversarial Cross-Scale Consistency Pursuit (ACSCP). On one hand, a U-net structured generation network is designed to generate density map from input patch, and an adversarial loss is directly employed to shrink the solution onto a realistic subspace, thus attenuating the blurry effects of density map estimation. On the other hand, we design a novel scale-consistency regularizer which enforces that the sum up of the crowd counts from local patches (i.e., small scale) is coherent with the overall count of their region union (i.e., large scale). The above losses are integrated via a joint training scheme, so as to help boost density estimation performance by further exploring the collaboration between both objectives. Extensive experiments on four benchmarks have well demonstrated the effectiveness of the proposed innovations as well as the superior performance over prior art.

Despite recent emergence of adversarial based methods for video prediction, existing algorithms often produce unsatisfied results in image regions with rich structural information (i.e., object boundary) and detailed motion (i.e., articulated body movement). To this end, we present a structure preserving video prediction framework to explicitly address above issues and enhance video prediction quality. On one hand, our framework contains a two-stream generation architecture which deals with high frequency video content (i.e., detailed object or articulated motion structure) and low frequency video content (i.e., location or moving directions) in two separate streams. On the other hand, we propose a RNN structure for video prediction, which employs temporal-adaptive convolutional kernels to capture time-varying motion patterns as well as tiny objects within a scene. Extensive experiments on diverse scenes, ranging from human motion to semantic layout prediction, demonstrate the effectiveness of the proposed video prediction approach.

Most human activity analysis works (i.e., recognition or prediction) only focus on a single granularity, i.e., either modelling global motion based on the coarse level movement such as human trajectories or forecasting future detailed action based on body parts' movement such as skeleton motion. In contrast, in this work, we propose a multi-granularity interaction prediction network which integrates both global motion and detailed local action. Built on a bidirectional LSTM network, the proposed method possesses between granularities links which encourage feature sharing as well as cross-feature consistency between both global and local granularity (e.g., trajectory or local action), and in turn predict long-term global location and local dynamics of each individual. We validate our method on several public datasets with promising performance.

Despite recent emergence of video caption methods, how to generate fine-grained video descriptions (i.e., long and detailed commentary about individual movements of multiple subjects as well as their frequent interactions) is far from being solved, which however has great applications such as automatic sports narrative. To this end, this work makes the following contributions. First, to facilitate this novel research of fine-grained video caption, we collected a novel dataset called Fine-grained Sports Narrative dataset (FSN) that contains 2K sports videos with ground-truth narratives from YouTube.com. Second, we develop a novel performance evaluation metric named Fine-grained Captioning Evaluation (FCE) to cope with this novel task. Considered as an extension of the widely used METEOR, it measures not only the linguistic performance but also whether the action details and their temporal orders are correctly described. Third, we propose a new framework for fine-grained sports narrative task. This network features three branches: 1) a spatio-temporal entity localization and role discovering sub-network; 2) a fine-grained action modeling sub-network for local skeleton motion description; and 3) a group relationship modeling sub-network to model interactions between players. We further fuse the features and decode them into long narratives by a hierarchically recurrent structure. Extensive experiments on the FSN dataset demonstrates the validity of the proposed framework for fine-grained video caption.

Scale problem lies in the heart of object detection. In this work, we develop a novel Scale-Transferrable Detection Network (STDN) for detecting multi-scale objects in images. In contrast to previous methods that simply combine object predictions from multiple feature maps from different network depths, the proposed network is equipped with embedded super-resolution layers (named as scale-transfer layer/module in this work) to explicitly explore the interscale consistency nature across multiple detection scales. Scale-transfer module naturally fits the base network with little computational cost. This module is further integrated with a dense convolutional network (DenseNet) to yield a one-stage object detector. We evaluate our proposed architecture on PASCAL VOC 2007 and MS COCO benchmark tasks and STDN obtains significant improvements over the comparable state-of-the-art detection models.

How to effectively approximate real-valued parameters with binary codes plays a central role in neural network binarization. In this work, we reveal an important fact that binarizing different layers has a widely varied effect on the compression ratio of network and the loss of performance. Based on this fact, we propose a novel and flexible neural network binarization method by introducing the concept of layer-wise priority which binarizes parameters in inverse order of their layer depth. In each training step, our method selects a specific network layer, minimizes the discrepancy between the original real-valued weights and its binary approximations following block coordinate descent scheme. During the iteration of the above process, it is significant that we can flexibly decide whether to binarize the remaining floating layers or not and explore a trade-off between the loss of performance and the compression ratio of model. The resulting binary network is applied for efficient pedestrian detection. Experimental results on several benchmarks show that under the same compression ratio, the model compressed by our method achieves much lower miss rate and faster detection speed than those obtained by the state-of-the-art neural network binarization method.

Typically, cropped and aligned face images are required as the input of a face recognition model. In contrast, popular object detectors based on deep convolutional network usually locate and classify objects simultaneously, which eliminates redundant computation. This work presents a single-network model called Uniface network for simultaneous face detection, landmark localization and recognition. We develop a feature sharing infrastructure for seamlessly integrate both the detection/localization module and the recognition module. To facilitate large-scale end-to-end training, we propose a method by encouraging top-level features of our model to mimic those of a well-trained single-task face recognition model. Comprehensive experiments on face detection, landmark localization and verification tasks demonstrate that the proposed network achieves competing performance in both face recognition benchmark (99.0% on LFW for a single model) and face detection benchmark (86.4% against 2000 false positives on FDDB for a single model).

We aim to automatically generate human motion sequence from a single input person image, with some specific action label. To this end, we propose a cross-space human motion video generation network which features two paths: a forward path that first samples/generates a sequence of low dimensional motion vectors based on Gaussian Process (GP), which is paired with the input person image to form a moving human figure sequence; and a backward path based on the predicted human images to re-extract the corresponding latent motion representations. As lack of supervision, the reconstructed latent motion representations are expected to be as close as possible to the GP sampled ones, thus yielding a cyclic objective function for cross-space (i.e., motion and appearance) mutual constrained generation. We further propose an alternative sampling/generation algorithm with respect to constraints from both spaces. Extensive experimental results show that the proposed framework successfully generates novel human motion sequences with reasonable visual quality.

State-of-the-art live face verification methods would easily be attacked by recorded facial expression sequence. This work directly addresses this issue via proposing a patch-wise motion parameterization based verification network infrastructure. This method directly explores the underlying subtle motion difference between the facial movements re-captured from a planer screen (e.g., a pad) and those from a real face; therefore interactive facial expression is no longer required. Furthermore, inspired by the fact that ?a fake facial movement sequence MUST contains many patch-wise fake sequences?, we embed our network into a multiple instance learning framework, which further enhance the recall rate of the proposed technique. Extensive experimental results on several face benchmarks well demonstrate the superior performance of our method.

Key to automatically generate natural scene images is to properly arrange amongst various spatial elements, especially in the depth cue. To this end, we introduce a novel depth structure preserving scene image generation network (DSP-GAN), which favors a hierarchical architecture, for the purpose of depth structure preserving scene image generation. The main trunk of the proposed infrastructure is built upon a Hawkes point process that models high-order spatial dependency between different depth layers. Within each layer generative adversarial sub-networks are trained collaboratively to generate realistic scene components, conditioned on the layer information produced by the point process. We experiment our model on annotated natural scene images collected from SUN dataset and demonstrate that our models are capable of generating depth-realistic natural scene image.

Recent style transfer methods based on deep networks strive to generate more content matching stylized images by adding semantic guidance in the iterative process. However, these approaches can just guarantee the transfer of integral color and texture distribution between semantically equivalent regions, but local variation within these regions cannot be accurately captured. Therefore, the resulting image lacks local plausibility. To this end, we develop a non-parametric patch based style transfer framework to synthesize more content coherent images. By designing a novel patch matching algorithm which simultaneously takes high-level category information and geometric structure information (e.g., human pose and building structure) into account, our proposed method is capable of transferring more detailed distribution and producing more photorealistic stylized images. We show that our approach achieves remarkable style transfer results on contents with geometric structure, including human body, vehicles, buildings, etc.

Most adversarial learning based video prediction methods suffer from image blur, since the commonly used adversarial and regression loss pair work rather in a competitive way than collaboration, yielding compromised blur effect. In the meantime, as often relying on a single-pass architecture, the predictor is inadequate to explicitly capture the forthcoming uncertainty. Our work involves two key insights: (1) Video prediction can be approached as a stochastic process: we sample a collection of proposals conforming to possible frame distribution at following time stamp, and one can select the final prediction from it. (2) De-coupling combined loss functions into dedicatedly designed sub-networks encourages them to work in a collaborative way. Combining above two insights we propose a two-stage network called VPSS (\textbf{V}ideo \textbf{P}rediction via \textbf{S}elective \textbf{S}ampling). Specifically a \emph{Sampling} module produces a collection of high quality proposals, facilitated by a multiple choice adversarial learning scheme, yielding diverse frame proposal set. Subsequently a \emph{Selection} module selects high possibility candidates from proposals and combines them to produce final prediction. Extensive experiments on diverse challenging datasets demonstrate the effectiveness of proposed video prediction approach, i.e., yielding more diverse proposals and accurate prediction results.

Most previous works on the first-person video recognition focus on measuring the similarity of different actions by using low-level features of objects interacting with humans. However, due to noisy camera motion and frequent changes in viewpoint and scale, they fail to capture and model highly discriminative object features. In this paper, we propose a novel pipeline for the first-person daily activity recognition. Our object feature extraction pipeline is inspired by the recent success of object hypotheses and deep convolutional neural network (CNN)-based detection frameworks. Our key contribution is a simple yet effective manipulated object proposal generation scheme. This scheme leverages motion cues, such as motion boundary and motion magnitude (in contrast, camera motion is usually considered as “noise” for most previous methods), to generate a more compact and discriminative set of object proposals, which are more closely related to the objects, which are being manipulated. Then, we learn more discriminative object detectors from these manipulated object proposals based on region-based CNN. Meanwhile, we develop a non-linear feature fusion scheme, which better combines object and motion features. We show in experiments that the proposed framework significantly outperforms the state-of-the-art recognition performance on a challenging first-person daily activity benchmark.

A fundamental bottleneck for achieving highly discriminative action representation is that local motion/appearance features are usually not semantic aligned. Namely, a local feature, such as a motion vector or motion trajectory, does not possess any attribute that indicates which moving body part or operated object it is associated with. This mostly leads to global feature pooling/representation learning methods that are often too coarse. Inspired by the recent success of end-to-end (pixel-to-pixel) deep convolutional neural networks (DCNNs), in this paper, we first propose a DCNN architecture, which maps a human centric image region onto human body part response maps. Based on these response maps, we propose a second DCNN, which achieves semantic-aligned feature representation learning. Prior knowledge that only a few parts are responsible for a certain action is also utilized by introducing a group (part) sparseness prior during feature learning. The learned semantic-aligned feature not only boosts the discriminative capability of action representation, but also possesses the good nature of robustness to pose variations and occlusions. Finally, an iterative mining method is employed for learning discriminative action primitive detectors. Extensive experiments on action recognition benchmarks demonstrate a superior recognition performance of the proposed framework.

Recent work has shown that optical flow estimation can be formulated as a supervised learning problem. Moreover, convolutional networks have been successfully applied to this task. However, supervised flow learning is obfuscated by the shortage of labeled training data. As a consequence, existing methods have to turn to large synthetic datasets for easily computer generated ground truth. In this work, we explore if a deep network for flow estimation can be trained without supervision. Using image warping by the estimated flow, we devise a simple yet effective unsupervised method for learning optical flow, by directly minimizing photometric consistency. We demonstrate that a flow network can be trained from end-to-end using our unsupervised scheme. In some cases, our results come tantalizingly close to the performance of methods trained with full supervision.

Traditional action recognition methods aim to recognize actions with complete observations/executions. However, it is often difficult to capture fully executed actions due to occlusions, interruptions, etc. Meanwhile, action prediction/recognition in advance based on partial observations is essential for preventing the situation from deteriorating. Besides, fast spotting human activities using partially observed data is a critical ingredient for retrieval systems. Inspired by the recent success of data binarization in efficient retrieval/recognition, we propose a novel approach, named Partial Reconstructive Binary Coding (PRBC), for action analysis based on limited frame glimpses during any period of the complete execution. Specifically, we learn discriminative compact binary codes for partial actions via a joint learning framework, which collaboratively tackles feature reconstruction as well as binary coding. We obtain the solution to PRBC based on a discrete alternating iteration algorithm. Extensive experiments on four realistic action datasets in terms of three tasks (i.e., partial action retrieval, recognition and prediction) clearly show the superiority of PRBC over the state-of-the-art methods, along with significantly reduced memory load and computational costs during the online test.

Recently, zero-shot action recognition (ZSAR) has emerged with the explosive growth of action categories. In this paper, we explore ZSAR from a novel perspective by adopting the Error-Correcting Output Codes (dubbed ZSECOC). Our ZSECOC equips the conventional ECOC with the additional capability of ZSAR, by addressing the domain shift problem. In particular, we learn discriminative ZSECOC for seen categories from both category-level semantics and intrinsic data structures. This procedure deals with domain shift implicitly by transferring the well-established correlations among seen categories to unseen ones. Moreover, a simple semantic transfer strategy is developed for explicitly transforming the learned embeddings of seen categories to better fit the underlying structure of unseen categories. As a consequence, our ZSECOC inherits the promising characteristics from ECOC as well as overcomes domain shift, making it more discriminative for ZSAR. We systematically evaluate ZSECOC on three realistic action benchmarks, i.e. Olympic Sports, HMDB51 and UCF101. The experimental results clearly show the superiority of ZSECOC over the state-of-the-art methods.

Modeling of high order interactional context, e.g., group interaction, lies in the central of collective/group activity recognition. However, most of the previous activity recognition methods do not offer a flexible and scalable scheme to handle the high order context modeling problem. To explicitly address this fundamental bottleneck, we propose a recurrent interactional context modeling scheme based on LSTM network. By utilizing the information propagation/aggregation capability of LSTM, the proposed scheme unifies the interactional feature modeling process for single person dynamics, intra-group (e.g., persons within a group) and inter-group(e.g., group to group)interactions. The proposed high order context modeling scheme produces more discriminative/descriptive interactional features. It is very flexible to handle a varying number of input instances (e.g., different number of persons in a group or different number of groups) and linearly scalable to high order context modeling problem. Extensive experiments on two benchmark collective/group activity datasets demonstrate the effectiveness of the proposed method.

We introduce a Multiple Granularity Analysis framework for video segmentation in a coarse-to-fine manner. We cast video segmentation as a spatio-temporal superpixel labeling problem. Benefited from the bounding volume provided by off-the-shelf object trackers, we estimate the foreground/ background super-pixel labeling using the spatiotemporal multiple instance learning algorithm to obtain coarse foreground/background separation within the volume. We further refine the segmentation mask in the pixel level using the graph-cut model. Extensive experiments on benchmark video datasets demonstrate the superior performance of the proposed video segmentation algorithm.

Input binarization has shown to be an effective way for network acceleration. However, previous binarization scheme could be regarded as simple pixel-wise thresholding operations (i.e., order-one approximation) and suffers a big accuracy loss. In this paper, we propose a high-order binarization scheme, which achieves more accurate approximation while still possesses the advantage of binary operation. In particular, the proposed scheme recursively performs residual quantization and yields a series of binary input images with decreasing magnitude scales. Accordingly, we propose high-order binary filtering and gradient propagation operations for both forward and backward computations. Theoretical analysis shows approximation error guarantee property of proposed method. Extensive experimental results demonstrate that the proposed scheme yields great recognition accuracy while being accelerated.

Most existing matching algorithms are one-off algorithms, i.e., they usually measure the distance between the two image feature representation vectors for only one time. In contrast, human's vision system achieves this task, i.e., image matching, by recursively looking at specific/related parts of both images and then making the final judgement. Towards this end, we propose a novel loopy recurrent neural network (Loopy RNN), which is capable of aggregating relationship information of two input images in a progressive/iterative manner and outputting the consolidated matching score in the final iteration. A Loopy RNN features two uniqueness. First, built on conventional long short-term memory (LSTM) nodes, it links the output gate of the tail node to the input gate of the head node, thus it brings up symmetry property required for matching. Second, a monotonous loss designed for the proposed network guarantees increasing confidence during the recursive matching process. Extensive experiments on several image matching benchmarks demonstrate the great potential of the proposed method.

Predicting human interaction is challenging as the on-going activity has to be inferred based on a partially observed video. Essentially, a good algorithm should effectively model the mutual influence between the two interacting subjects. Also, only a small region in the scene is discriminative for identifying the on-going interaction. In this work, we propose a relative attention model to explicitly address these difficulties. Built on a tri-coupled deep recurrent structure representing both interacting subjects and global interaction status, the proposed network collects spatio-temporal information from each subject, rectified with global interaction information, yielding effective interaction representation. Moreover, the proposed network also unifies an attention module to assign higher importance to the regions which are relevant to the on-going action. Extensive experiments have been conducted on two public datasets, and the results demonstrate that the proposed relative attention network successfully predicts informative regions between interacting subjects, which in turn yields superior human interaction prediction accuracy.

This paper proposes a novel recursive hashing scheme, in contrast to conventional "one-off" based hashing algorithms. Inspired by human's "nonsalient-to-salient" perception path, the proposed hashing scheme generates a series of binary codes based on progressively expanded salient regions. Built on a recurrent deep network, i.e., LSTM structure, the binary codes generated from later output nodes naturally inherit information aggregated from previously codes while explore novel information from the extended salient region, and therefore it possesses good scalability property. The proposed deep hashing network is trained via minimizing a triplet ranking loss, which is end-to-end trainable. Extensive experimental results on several image retrieval benchmarks demonstrate good performance gain over state-of-the-art image retrieval methods and its scalability property.

Scale analysis plays a vital role in pedestrian detection. Conventional approaches usually directly concatenate multi-scale outputs, which is only capable of modeling first-order dependency among various scales. In contrast, this work proposes a novel scale-context modeling scheme by exploiting the highly nonlinear dependency among scales. The proposed scheme aggregates output response maps from mid-results of convolutional layers via a bi-directional recurrent sub-network. Therefore scale information could flow among different layers and implicit underlying dependency structure information in the scale space would be disclosed, which yields more consistency detection. Experimental results on Caltech Pedestrian detection benchmark demonstrate the superior detection (state-of-the-art miss rate of 8.56%) of the proposed method over prior art.

Multi-shot person Re-IDentification (Re-ID) has recently received more research attention as its problem setting is more realistic compared to single-shot Re-ID in terms of application. While many large-scale single-shot Re-ID human image datasets have been released, most existing multishot Re-ID video sequence datasets containonly a few (i.e., several hundreds) human instances, which hinders further improvement of multi-shot Re-ID performance. To this end, we propose a deep cross-modality alignment network, which jointly explores both human sequence pairs and image pairs to facilitate training better multi-shot human Re-ID models, i.e., via transferring knowledge from image data to sequence data. To mitigate modality-to-modality mismatch issue, the proposed network is equipped with an image-to-sequence adaption module called cross-modality alignment sub-network, which successfully maps each human image into a pseudo human sequence to facilitate knowledge transferring and joint training. Extensive experimental results on several multi-shot person Re-ID benchmarks demonstrate great performance gain brought up by the proposed network.

Fine-grained object recognition is challenging due to large intra-class variation and inter-class ambiguity. A good algorithm should be able to: 1) discover discriminative local details and 2) align and aggregate these local discriminative patch-level features in an effective way to facilitate object level classification. Towards this end, we propose a novel local feature discovery, discriminative alignment and aggregation framework, inspired by the recent success of deep recurrent attention model. First, we develop a novel attribute-guided attentive network to sequentially discover informative parts/regions, by seeking a good registration between attentive regions and predefined object attributes. This could be considered as a semantic guided salient region discovery and alignment network, which might be more robust than conventional attention model. Second, these discovered regions are actively and progressively fed into a recurrent neural network, to yield the object-level representation. This could be considered as a discriminant aggregation network and informative patch-level features are propagated and accumulated to the deeper nodes of the recurrent network for final classification. We extensively test our framework on two fine-grained image benchmarks and the results demonstrate the effectiveness of the proposed framework.

This work makes the first attempt to generate articulated human motion sequence from a single image. On one hand, we utilize paired inputs including human skeleton information as motion embedding and a single human image as appearance reference, to generate novel motion frames based on the conditional GAN infrastructure. On the other hand, a triplet loss is employed to pursue appearance smoothness between consecutive frames. As the proposed framework is capable of jointly exploiting the image appearance space and articulated/kinematic motion space, it generates realistic articulated motion sequence, in contrast to most previous video generation methods which yield blurred motion effects. We test our model on two human action datasets including KTH and Human3.6M, and the proposed framework generates very promising results on both datasets.

Single-image rain removal has always been challenging, due to its inherent ill-posed nature. In this paper, we propose a novel rain removal pipeline. This pipeline features two processing pathways: 1) a bottom-up pathway which iteratively performs down sampling and subtraction operations, yielding a Laplacian pyramid (LP) with decreasing scales, to bypass rain details; and 2) a top-down pathway which is built on image series obtained by the above bottom-up process and generate a series of image components based on conditional generative adversarial nets (CGAN), aiming at removing rain streaks while preserving image details. Generated components are further fused via scale-space aggregation, and descriptive image intrinsic information contained in different scales is interacted to jointly generate high quality de-rained image. Extensive experiments on synthetic and real-world rainy image datasets both demonstrate the effectiveness of the proposed rain streaks removal approach.

Lack of large available datasets fully annotated is a fundamental bottleneck in pulmonary nodule detection, especially when the sensing equipment and the corresponding computed tomography (CT) images obtained are device dependent. This work presents a novel cross modal scheme, pursuing modal alignment, to facilitate our aggregate channel detector training. Named as multi-class cycle-consistent adversarial network (CycleGAN), our proposed framework utilizes a generative adversarial model to transfer nodule morphological characteristics from source modal to target modal, and we propose an end to end objective function to unify the transfer and detection procedures. The outputs of the two parts are combined with a dedicated fusion method for final classification. Extensive experimental results on 1948 scans of the private dataset demonstrate the proposed modal transfer method is very effective in data augmentation.

Fine grained video action analysis often requires reliable detection and tracking of various interacting objects and human body parts, denoted as Interactional Object Parsing. However, most of the previous methods based on either independent or joint object detection might suffer from high model complexity and challenging image content, e.g., illumination/pose/appearance/scale variation, motion, and occlusion etc. In this work, we propose an end-to-end system based on recurrent neural network to perform frame by frame interactional object parsing, which can alleviate the difficulty through an incremental/progressive manner. Our key innovation is that: instead of jointly outputting all object detections at once, for each frame we use a set of long-short term memory (LSTM) nodes to incrementally refine the detections. After passing through each LSTM node, more object detections are consolidated and thus more contextual information could be utilized to localize more difficult objects. The object parsing results are further utilized to form object specific action representation for fine grained action detection. Extensive experiments on two benchmark fine grained activity datasets demonstrate that our proposed algorithm achieves better interacting object detection performance, which in turn boosts the action recognition performance over the state-of-the-art.

We investigate the feature design and classification architectures in temporal action localization. This application focuses on detecting and labeling actions in untrimmed videos, which brings more challenge than classifying presegmented videos. The major difficulty for action localization is the uncertainty of action occurrence and utilization of information from different scales. Two innovations are proposed to address this issue. First, we propose a Pyramid of Score Distribution Feature (PSDF) to capture the motion information at multiple resolutions centered at each detection window. This novel feature mitigates the influence of unknown action position and duration, and shows significant performance gain over previous detection approaches. Second, inter-frame consistency is further explored by incorporating PSDF into the state-of-the-art Recurrent Neural Networks, which gives additional performance gain in detecting actions in temporally untrimmed videos. We tested our action localization framework on the THUMOS'15 and MPII Cooking Activities Dataset, both of which show a large performance improvement over previous attempts.

Knowing how hands move and what object is being manipulated are two key sub-tasks for analyzing first-person (egocentric) action. However, lack of fully annotated hand data as well as imprecise foreground segmentation make either sub-task challenging. This work aims to explicitly ad dress these two issues via introducing a cascaded interactional targeting (i.e., infer both hand and active object regions) deep neural network. Firstly, a novel EM-like learning framework is proposed to train the pixel-level deep convolutional neural network (DCNN) by seamlessly integrating weakly supervised data (i.e., massive bounding box annotations) with a small set of strongly supervised data (i.e., fully annotated hand segmentation maps) to achieve state-of-the-art hand segmentation performance. Secondly, the resulting high-quality hand segmentation maps are further paired with the corresponding motion maps and object feature maps, in order to explore the contextual information among object, motion and hand to generate interactional foreground regions (operated objects). The resulting interactional target maps (hand + active object) from our cascaded DCNN are further utilized to form discriminative action representation. Experiments show that our framework has achieved the state-of-the-art egocentric action recognition performance on the benchmark dataset Activities of Daily Living (ADL).

Convolutional Neural Network (CNN) has demonstrated promising performance in single-label image classification tasks. However, how CNN best copes with multi-label images still remains an open problem, mainly due to the complex underlying object layouts and insufficient multi-label training images. In this work, we propose a flexible deep CNN infrastructure, called Hypotheses-CNN-Pooling (HCP), where an arbitrary number of object segment hypotheses are taken as the inputs, then a shared CNN is connected with each hypothesis, and finally the CNN output results from different hypotheses are aggregated with max pooling to produce the ultimate multi-label predictions. Some unique characteristics of this flexible deep CNN infrastructure include: 1) no ground-truth bounding box information is required for training; 2) the whole HCP infrastructure is robust to possibly noisy and/or redundant hypotheses; 3) the shared CNN is flexible and can be well pre-trained with a large-scale single-label image dataset, e.g., ImageNet; and 4) it may naturally output multi-label prediction results. Experimental results on Pascal VOC 2007 and VOC 2012 multi-label image datasets well demonstrate the superiority of the proposed HCP infrastructure over other state-of-the-arts. In particular, the mAP reaches 90.5% by HCP only and 93.2% after the fusion with our complementary result in [44] based on hand-crafted features on the VOC 2012 dataset.

Correlation filters have been widely applied to visual tracking in recent years as adaptive correlation filters with short-term memory are robust to large appearance changes. However, tracking methods relying on correlation filters are prone to drifting due to noisy updates. Moreover, these methods are unable to recover from tracking failures caused by temporary or persistent heavy occlusions. In this paper, we interpret correlation filters as the counterparts of convolution filters in deep neural networks. Correlation filters encode the holistic template of target appearance, while convolution filters with smaller size encode the part-based template. In the light of this idea, we propose to exploit deep convolutional networks that directly learn mapping as a spatial correlation between two consecutive frames for visual tracking. We show that these deeply learned networks are effective in maintaining the long-term memory of target appearance for handling heavy occlusion or out-of-view. We further take the response maps both from the deep networks and conventional correlation filters into account for precisely locating the target. Experimental results on large-scale benchmark sequences show that the proposed algorithm performs favorably against the state-of-the-art methods.

This article investigates a low-rank representation--based graph, which can used in graph-based vision tasks including image segmentation and image annotation. It naturally fuses multiple types of image features in a framework named multitask low-rank affinity pursuit. Given the image patches described with multiple types of features, we aim at inferring a unified affinity matrix that implicitly encodes the relations among these patches. This is achieved by seeking the sparsity-consistent low-rank affinities from the joint decompositions of multiple feature matrices into pairs of sparse and low-rank matrices, the latter of which is expressed as the production of the image feature matrix and its corresponding image affinity matrix. The inference process is formulated as a minimization problem and solved efficiently with the augmented Lagrange multiplier method. Considering image patches as vertices, a graph can be built based on the resulted affinity matrix. Compared to previous methods, which are usually based on a single type of feature, the proposed method seamlessly integrates multiple types of features to jointly produce the affinity matrix in a single inference step. The proposed method is applied to graph-based image segmentation and graph-based image annotation. Experiments on benchmark datasets well validate the superiority of using multiple features over single feature and also the superiority of our method over conventional methods for feature fusion.

Feature learning is an intensively studied research topic in image classification. Although existing methods like sparse coding, locality-constrained linear coding, fisher vector encoding, etc., have shown their effectiveness in image representation, most of them overlook a phenomenon called thesmall sample size problem, where the number of training samples is relatively smaller than the dimensionality of the features, which may limit the predictive power of the classifier. Subspace learning is a strategy to mitigate this problem by reducing the dimensionality of the features. However, most conventional subspace learning methods attempt to learn a global subspace to discriminate all the classes, which proves to be difficult and ineffective in multi-class classification task. To this end, we propose to learn a local subspace for each sample instead of learning a global subspace for all samples. Our key observation is that, in multi-class image classification, the label of each testing sample is only confused by a few classes which have very similar visual appearance to it. Thus, in this work, we propose a coarse-to-fine strategy, which first picks out such classes, and then conducts a local subspace learning to discriminate them. As the subspace learning method is regularized and conducted within some selected classes, we term it selective regularized subspace learning (SRSL), and we term our classification pipeline selective regularized subspace learning based multi-class image classification (SRSL_MIC). Experimental results on four representative datasets (Caltech-101, Indoor-67, ORL Faces and AR Faces) demonstrate the effectiveness of the proposed method.

Dense local trajectories have been successfully used in action recognition. However, for most actions only a few local motion features (e.g., critical movement of hand, arm, leg etc.) are responsible for the action label. Therefore, highlighting the local features which are associated with important motion parts will lead to a more discriminative action representation. Inspired by recent advances in sentence regularization for text classification, we introduce a Motion Part Regularization framework to mine for discriminative groups of dense trajectories which form important motion parts. First, motion part candidates are generated by spatio-temporal grouping of densely extracted trajectories. Second, an objective function which encourages sparse selection for these trajectory groups is formulated together with an action class discriminative term. Then, we propose an alternative optimization algorithm to efficiently solve this objective function by introducing a set of auxiliary variables which correspond to the discriminativeness weights of each motion part (trajectory group). These learned motion part weights are further utilized to form a discriminativeness weighted Fisher vector representation for each action sample for final classification. The proposed motion part regularization framework achieves the state-of-the-art performances on several action recognition benchmarks.

Modeling human-object interactions and manipulating motions lies in the heart of fine-grained action recognition. Previous methods heavily rely on explicit detection of the object being interacted, which requires intensive human labour on object annotation. To bypass this constraint and achieve better classification performance, in this work, we propose a novel fine-grained action recognition pipeline by interaction part proposal and discriminative mid-level part mining. Firstly, we generate a large number of candidate object regions using off-the-shelf object proposal tool, e.g., BING. Secondly, these object regions are matched and tracked across frames to form a large spatio-temporal graph based on the appearance matching and the dense motion trajectories through them. We then propose an efficient approximate graph segmentation algorithm to partition and filter the graph into consistent local dense sub-graphs. These sub-graphs, which are spatio-temporal sub-volumes, represent our candidate interaction parts. Finally, we mine discriminative mid-level part detectors from the features computed over the candidate interaction parts. Bag-of-detection scores based on a novel Max-N pooling scheme are computed as the action representation for a video sample. We conduct extensive experiments on human-object interaction datasets including MPII Cooking and MSR Daily Activity 3D. The experimental results demonstrate that the proposed framework achieves consistent improvements over the state-of-the-art action recognition accuracies on the benchmarks, without using any object annotation.

We present LiveTraj, a novel system for tracking trajectories in a live video stream in real time, backed by a cloud platform. Although trajectory tracking is a well-studied topic in computer vision, so far most attention has been devoted to improving the accuracy of trajectory tracking, rather than the efficiency. To our knowledge, LiveTraj is the first that achieves real-time efficiency in trajectory tracking, which can be a key enabler in many important applications such as video surveillance, action recognition and robotics. LiveTraj is based on a state-of-the-art approach to (offline) trajectory tracking; its main innovation is to adapt this base solution to run on an elastic cloud platform to achieve real-time tracking speed at an affordable cost. The video demo shows the offline base solution and LiveTraj side by side, both running on a video stream containing human actions. Besides demonstrating the real-time efficiency of LiveTraj, our video demo also exhibits important system parameters to the audience such as latency and cloud resource usage for different components of the system. Further, if the conference venue provides sufficiently fast Internet connection to our cloud platform, we also plan to demonstrate LiveTraj on-site, during which we will show LiveTraj identifying and tracking trajectories from a live video stream captured by a camera.

This article summarizes the corresponding half-day tutorial at ACM Multimedia 2015. This tutorial reviews recent progresses in human-centric images and videos analysis: 1) fashion analysis: parsing, attribute prediction and retrieval; 2) action analysis: discriminative feature selection, pooling and fusion; 3) person verification: cross-domain person verification via learning a generalized similarity measure, and bit-scalable deep hashing with regularized similarity learning.

In this paper, we investigate order-preserving sparse coding for classifying structured data whose atomic features possess ordering relationships. Examples include time sequences where individual frame-wise features are temporally ordered, as well as still images (landscape, street view, etc.) where different regions of the image are spatially ordered. Classification of these structured data is often tackled by first decomposing the input data into individual atomic features, then performing sparse coding or other processing for each atomic feature vector independently, and finally aggregating individual responses to classify the input data. However, this heuristic approach ignores the underlying order of the individual atomic features within the input data, and results in suboptimal discriminative capability. In this work, we introduce an order preserving regularizer which aims to preserve the ordering structure of the reconstruction coefficients within the sparse coding framework. An efficient Nesterov-type smooth approximation method is developed for optimization of the new regularization criterion, with theoretically guaranteed error bound. We perform extensive experiments for time series classification on a synthetic dataset, several machine learning benchmarks, and an RGB-D human activity dataset. We also report experiments for scene classification on a benchmark image dataset. The encoded representation is discriminative and robust, and our classifier outperforms state-of-the-art methods on these tasks.

Automated scene analysis has been a topic of great interest in computer vision and cognitive science. Recently, with the growth of crowd phenomena in the real world, crowded scene analysis has attracted much attention. However, the visual occlusions and ambiguities in crowded scenes, as well as the complex behaviors and scene semantics, make the analysis a challenging task. In the past few years, an increasing number of works on the crowded scene analysis have been reported, which covered different aspects including crowd motion pattern learning, crowd behavior and activity analyses, and anomaly detection in crowds. This paper surveys the state-of-the-art techniques on this topic. We first provide the background knowledge and the available features related to crowded scenes. Then, existing models, popular algorithms, evaluation protocols, and system performance are provided corresponding to different aspects of the crowded scene analysis. We also outline the available datasets for performance evaluation. Finally, some research problems and promising future directions are presented with discussions.

In this paper, a novel system is developed for synthesizing user-specified emotions onto arbitrary input images or videos. Other than defining the visual affective model based on empirical knowledge, a data-driven learning framework is proposed to extract the emotion-related knowledge from a set of emotion-annotated images. In a divide-and-conquer manner, the images are clustered into several emotion-specific scene subgroups for model learning. The visual affection is modeled with Gaussian mixture models based on color features of local image patches. For the purpose of affective filtering, the feature distribution of the target is aligned to the statistical model constructed from the emotion-specific scene subgroup, through a piecewise linear transformation. The transformation is derived through a learning algorithm, which is developed with the incorporation of a regularization term enforcing spatial smoothness, edge preservation, and temporal smoothness for the derived image or video transformation. Optimization of the objective function is sought via standard nonlinear method. Intensive experimental results and user studies demonstrate that the proposed affective filtering framework can yield effective and natural effects for images and videos.

This paper introduces a novel approach to facilitating image search based on a compact semantic embedding. A novel method is developed to explicitly map concepts and image contents into a unified latent semantic space for the representation of semantic concept prototypes. Then, a linear embedding matrix is learned that maps images into the semantic space, such that each image is closer to its relevant concept prototype than other prototypes. In our approach, the semantic concepts equated with query keywords and the images mapped into the vicinity of the prototype are retrieved by our scheme. In addition, a computationally efficient method is introduced to incorporate new semantic concept prototypes into the semantic space by updating the embedding matrix. This novelty improves the scalability of the method and allows it to be applied to dynamic image repositories. Therefore, the proposed approach not only narrows semantic gap but also supports an efficient image search process. We have carried out extensive experiments on various cross-modality image search tasks over three widely-used benchmark image datasets. Results demonstrate the superior effectiveness, efficiency, and scalability of our proposed approach.

In this paper, we propose a novel approach to segment moving object in video by utilizing improved point trajectories . First, point trajectories are densely sampled from video and tracked through optical flow, which provides information of long-term temporal interactions among objects in the video sequence . Second, a novel affinity measurement method considering both global and local information of point trajectories is proposed to cluster trajectories into groups. Finally, we propose a new graph-based segmentation method which adopts both local and global motion information encoded by the tracked dense point trajectories. The proposed approach achieves good performance on trajectory clustering, and it also obtains accurate video object segmentation results on both the Moseg dataset and our new dataset containing more challenging videos.

In kernel based learning, the kernel trick transforms the original representation of a feature instance into a vector of similarities with the training feature instances, known as kernel representation. However, feature instances are sometimes ambiguous and the kernel representation calculated based on them do not possess any discriminative information, which can eventually harm the trained classifier. To address this issue, we propose to automatically select good feature instances when calculating the kernel representation in multiple kernel learning. Specifically, for the kernel representation calculated for each input feature instance, we multiply it element-wise with a latent binary vector named as instance selection variables, which targets at selecting good instances and attenuate the effect of ambiguous ones in the resulting new kernel representation. Beta process is employed for generating the prior distribution for the latent instance selection variables. We then propose a Bayesian graphical model which integrates both MKL learning and inference for the distribution of the latent instance selection variables. Variational inference is derived for model learning under a max-margin principle. Our method is called Beta process multiple kernel learning. Extensive experiments demonstrate the effectiveness of our method on instance selection and its high discriminative capability for various classification problems in vision.

We propose to decompose the fine-grained human activ- ity analysis problem into two sequential tasks with increas- ing granularity. Firstly, we infer the coarse interaction sta- tus, i.e., which object is being manipulated and where it is. Knowing that the major challenge is frequent mutual oc- clusions during manipulation, we propose an "interaction tracking" framework in which hand/object position and in- teraction status are jointly tracked by explicitly modeling the contextual information between mutual occlusion and interaction status. Secondly, the inferred hand/object posi- tion and interaction status are utilized to provide 1) more compact feature pooling by effectively pruning large num- ber of motion features from irrelevant spatio-temporal po- sitions and 2) discriminative action detection by a granu- larity fusion strategy. Comprehensive experiments on two challenging fine-grained activity datasets (i.e., cooking ac- tion) show that the proposed framework achieves high ac- curacy/robustness in tracking multiple mutually occluded hands/objects during manipulation as well as significant performance improvement on fine-grained action detection over state-of-the-art methods.

In this work, we propose an alternative ground truth to the eye fixation map in visual attention study, called touch saliency. As it can be directly collected from the recorded data of users' daily browsing behavior on widely used smart phone devices with touch screens, the touch saliency data is easy to obtain. Due to the limited screen size, smart phone users usually move and zoom in the images, and fix the region of interest on the screen when browsing images. Our studies are two-fold. First, we collect and study the characteristics of these touch screen fixation maps (named touch saliency) by comprehensive comparisons with their counterpart, the eye-fixation maps (namely, visual saliency). The comparisons show that the touch saliency is highly correlated with the eye fixations for the same stimuli, which indicates its utility in data collection for visual attention study. Based on the consistency between both touch saliency and visual saliency, our second task is to propose a unified saliency prediction model for both visual and touch saliency detection. This model utilizes middle-level object category features extracted from pre-segmented image superpixels as input to the recently proposed multitask sparsity pursuit (MTSP) framework for saliency prediction. Extensive evaluations show that the proposed middle-level category features can considerably improve the saliency prediction performance when taking both touch saliency and visual saliency as ground truth.

We aim to unsupervisedly discover human's action (motion) patterns of manipulating various objects in scenarios such as assisted living. We are motivated by two key observations. First, large variation exists in motion patterns associated with various types of objects being manipulated, thus manually defining motion primitives is infeasible. Second, some motion patterns are shared among different objects being manipulated while others are object specific. We therefore propose a nonparametric Bayesian method that adopts a hierarchical Dirichlet process prior to learn representative manipulation (motion) patterns in an unsupervised manner. Taking easy-to-obtain object detection score maps and dense motion trajectories as inputs, the proposed probabilistic model can discover motion pattern groups associated with different types of objects being manipulated with a shared manipulation pattern dictionary. The size of the learned dictionary is automatically inferred. Comprehensive experiments on two assisted living benchmarks and a cooking motion dataset demonstrate superiority of our learned manipulation pattern dictionary in representing manipulation actions for recognition.

Recognizing complex human activities usually requires the detection and modeling of individual visual features and the interactions between them. Current methods only rely on the visual features extracted from 2-D images, and therefore often lead to unreliable salient visual feature detection and inaccurate modeling of the interaction context between individual features. In this paper, we show that these problems can be addressed by combining data from a conventional camera and a depth sensor (e.g., Microsoft Kinect). We propose a novel complex activity recognition and localization framework that effectively fuses information from both grayscale and depth image channels at multiple levels of the video processing pipeline. In the individual visual feature detection level, depth-based filters are applied to the detected human/object rectangles to remove false detections. In the next level of interaction modeling, 3-D spatial and temporal contexts among human subjects or objects are extracted by integrating information from both grayscale and depth images. Depth information is also utilized to distinguish different types of indoor scenes. Finally, a latent structural model is developed to integrate the information from multiple levels of video processing for an activity detection. Extensive experiments on two activity recognition benchmarks (one with depth information) and a challenging grayscale + depth human activity database that contains complex interactions between human-human, human-object, and human-surroundings demonstrate the effectiveness of the proposed multilevel grayscale + depth fusion scheme. Higher recognition and localization accuracies are obtained relative to the previous methods.

In this paper, we propose a novel method for spatial context modeling toward boosting visual discriminating power. We are particularly interested in how to model high-order local spatial contexts instead of the intensively studied second-order spatial contexts, i.e., co-occurrence relations. Motivated by the recent success of random forest in learning discriminative visual codebook, we present a spatialized random forest (SRF) approach, which can encode an unlimited length of high-order local spatial contexts. By spatially random neighbor selection and random histogram-bin partition during the tree construction, the SRF can explore much more complicated and informative local spatial patterns in a randomized manner. Owing to the discriminative capability test for the random partition in each tree node's split process, a set of informative high-order local spatial patterns are derived, and new images are then encoded by counting the occurrences of such discriminative local spatial patterns. Extensive comparison experiments on face recognition and object/scene classification clearly demonstrate the superiority of the proposed spatial context modeling method over other state-of-the-art approaches for this purpose.

In this paper, we propose a computational framework, called Image Re-Attentionizing, to endow the target region in an image with the ability of attracting human visual attention. In particular, the objective is to recolor the target patches by color transfer with naturalness and smoothness preserved yet visual attention augmented. We propose to approach this objective within the Markov Random Field (MRF) framework and an extended graph cuts method is developed to pursue the solution. The input image is first over-segmented into patches, and the patches within the target region as well as their neighbors are used to construct the consistency graphs. Within the MRF framework, the unitary potentials are defined to encourage each target patch to match the patches with similar shapes and textures from a large salient patch database, each of which corresponds to a high-saliency region in one image, while the spatial and color coherence is reinforced as pairwise potentials. We evaluate the proposed method on the direct human fixation data. The results demonstrate that the target region(s) successfully attract human attention and in the meantime both spatial and color coherence is well preserved.

In this paper, we present an intelligent photography system which can recommend the most user-favored view rectangle for arbitrary camera input, from a photographic compositional perspective. Automating this process is difficult, due to the subjectivity of human's aesthetics judgement and large variations of image contents, where heuristic compositional rules lack generality. Motivated by the recent prevalence of photo-sharing websites, e.g., Flickr.com, we develop a learning-based framework which discovers the underlying aesthetic photographic compositional structures from a large set of user-favored online sharing photographs and utilizes the implicitly shared knowledge among the professional photographers for aesthetically optimal view recommendation. In particular, we propose an Omni-Range Context method which explicitly encodes the spatial and geometric distributions of various visual elements in the photograph as well as cooccurrence characteristics of visual element pairs by using generative mixture models. Searching the optimal view rectangle is then formulated as maximum a posterior by imposing the trained prior distributions along with additional photographic constraints. The proposed system has the potential to operate in near real-time. Comprehensive user studies well demonstrate the effectiveness of the proposed framework for aesthetically optimal view recommendation.

Decrypting the secret of beauty or attractiveness has been the pursuit of artists and philosophers for centuries. To date, the computational model for attractiveness estimation has been actively explored in computer vision and multimedia community, yet with the focus mainly on facial features. In this article, we conduct a comprehensive study on female attractiveness conveyed by single/multiple modalities of cues, that is, face, dressing and/or voice, and aim to discover how different modalities individually and collectively affect the human sense of beauty. To extensively investigate the problem, we collect the Multi-Modality Beauty (M2B) dataset, which is annotated with attractiveness levels converted from manualk-wise ratings and semantic attributes of different modalities. Inspired by the common consensus that middle-level attribute prediction can assist higher-level computer vision tasks, we manually labeled many attributes for each modality. Next, a tri-layer Dual-supervised Feature-Attribute-Task (DFAT) network is proposed to jointly learn the attribute model and attractiveness model of single/multiple modalities. To remedy possible loss of information caused by incomplete manual attributes, we also propose a novel Latent Dual-supervised Feature-Attribute-Task (LDFAT) network, where latent attributes are combined with manual attributes to contribute to the final attractiveness estimation. The extensive experimental evaluations on the collected M2B dataset well demonstrate the effectiveness of the proposed DFAT and LDFAT networks for female attractiveness prediction.

Spatial contexts encode rich discriminative information for visual classification. However, as object shapes and scales vary significantly among images, spatial contexts with manually specified distance ranges are not guaranteed with optimality. In this work, we investigate how to automatically select discriminative and stable distance bin groups for modeling image spatial contexts to improve classification performance. We make two observations. First, the number of distance bins for context modeling can be arbitrarily large, and discriminative contexts are only from a small subset of distance bins. Second, adjacent distance bins for contexts modeling often show similar characteristics, thus encouraging grouping them together can result in more stable representation. Utilizing these two observations, we propose an omni-range spatial context mining framework for image classification. A sparse selection and grouping regularizer is employed along with an empirical risk, to discover discriminative and stable distance bin groups for context modeling. To facilitate efficient optimization, the objective function is approximated by a smooth convex function with theoretically guaranteed error bounds. The selected and grouped image spatial contexts, which are applied in food and national flag recognition, are demonstrated to be discriminative, compact and robust.

In this work, we develop a computer vision based fall prevention system for hospital ward application. To prevent potential falls, once the event of patient get up from the bed is automatically detected, nursing staffs are alarmed immediately for assistance. For the detection task, we use a RGBD sensor (Microsoft Kinect). The geometric prior knowledge is exploited by identifying a set of task-specific feature channels, e.g., regions of interest. Extensive motion and shape features from both color and depth image sequences are extracted. Features from multiple modalities and channels are fused via a multiple kernel learning framework for training the event detector. Experimental results demonstrate the high accuracy and efficiency achieved by the proposed system.

Discovering the secret of beauty has been the pursuit of artists and philosophers for centuries. Nowadays, the computational model for beauty estimation has been actively explored in computer science community, yet with the focus mainly on facial features. In this work, we perform a comprehensive study of female attractiveness conveyed by single/multiple modalities of cues, i.e., face, dressing and/or voice, and aim to uncover how different modalities individually and collectively affect the human sense of beauty. To this end, we collect the first Multi-Modality Beauty (M2B) dataset in the world for female attractiveness study, which is thoroughly annotated with attractiveness levels converted from manual k-wise ratings and semantic attributes of different modalities. A novel Dual-supervised Feature-Attribute-Task (DFAT) network is proposed to jointly learn the beauty estimation models of single/multiple modalities as well as the attribute estimation models. The DFAT network differentiates itself by its supervision in both attribute and task layers. Several interesting beauty-sense observations over single/multiple modalities are reported, and the extensive experimental evaluations on the collected M2B dataset well demonstrate the effectiveness of the proposed DFAT network for female attractiveness estimation.

In this work, we propose a new concept of touch saliency, and attempt to answer the question of whether the underlying image saliency map may be implicitly derived from the accumulative touch behaviors (or more specifically speaking, zoom-in and panning manipulations) when many users browse the image on smart mobile devices with multi-touch display of small size. The touch saliency maps are collected for the images of the recently released NUSEF dataset, and the preliminary comparison study demonstrates: 1) the touch saliency map is highly correlated with human eye fixation map for the same stimuli, yet compared to the latter, the touch data collection is much more flexible and requires no cooperation from users; and 2) the touch saliency is also well predictable by popular saliency detection algorithms. This study opens a new research direction of multimedia analysis by harnessing human touch information on increasingly popular multi-touch smart mobile devices.

Histograms have been widely used for feature representation in image and video content analysis. However, due to the orderless nature of the summarization process, histograms generally lack spatial information. This may degrade their discrimination capability in visual classification tasks. Although there have been several research attempts to encode spatial context into histograms, how to extend the encodings to higher order spatial context is still an open problem. In this paper,we propose a general histogram contextualization method to encode efficiently higher order spatial context. The method is based on the cooccurrence of local visual homogeneity patterns and hence is able to generate more discriminative histogram representations while remaining compact and robust. Moreover, we also investigate how to extend the histogram contextualization to multiple modalities of context. It is shown that the proposed method can be naturally extended to combine both temporal and spatial context and facilitate video content analysis. In addition, a method to combine cross-feature context with spatial context via the technique of random forest is also introduced in this paper. Comprehensive experiments on face image classification and human activity recognition tasks demonstrate the superiority of the proposed histogram contextualization method compared with the existing encoding methods.

In this paper, we present a novel framework, called learning by propagability, for two essential data mining tasks, i.e., classification and regression. The whole learning process is driven by the philosophy that the data labels and the optimal feature representation jointly constitute a harmonic system, where the data labels are invariant with respect to the propagation on the similarity graph constructed based on the optimal feature representation. Based on this philosophy, a unified framework of learning by propagability is proposed for the purposes of both classification and regression. Specifically, this framework has three characteristics: 1) the formulation unifies the label propagation and optimal feature representation pursuing, and thus the label propagation process is enhanced by benefiting from the refined similarity graph constructed with the derived optimal feature representation instead of the original representation; 2) it unifies the formulations for supervised and semisupervised learning in both classification and regression tasks; and 3) it can directly deal with the multiclass classification problems. Extensive experiments for the classification task on UCI toy data sets, handwritten digit recognition, face recognition, and microarray recognition as well as for the regression task of human age estimation on the FG-NET aging database, all validate the effectiveness of our proposed learning framework, compared with the state-of-the-art counterparts.

Modern visual classification models generally include a feature pooling step, which aggregates local features over the region of interest into a statistic through a certain spatial pooling operation. Two commonly used operations are the average and max poolings. However, recent theoretical analysis has indicated that neither of these two pooling techniques may be qualified to be optimal. Besides, we further reveal in this work that more severe limitations of these two pooling methods are from the unrecoverable loss of the spatial information during the statistical summarization and the underlying over-simplified assumption about the feature distribution. We aim to address these inherent issues in this work and generalize previous pooling methods as follows. We define a weighted ℓp-norm spatial pooling function tailored for the class-specific feature spatial distribution. Moreover, a sensible prior for the feature spatial correlation is incorporated. Optimizing such pooling function towards optimal class separability yields a so-called geometric ℓp-norm pooling (GLP) method. The described GLP method is capable of preserving the class-specific spatial/geometric information in the pooled features and significantly boosts the discriminating capability of the resultant features for image classification. Comprehensive evaluations on several image benchmarks demonstrate that the proposed GLP method can boost the image classification performance with a single type of feature to outperform or be comparable with the state-of-the-arts.

Many existing techniques for analyzing face images assume that the faces are at nearly frontal. Generalizing to non-frontal faces is often difficult, due to a dearth of ground truth for non-frontal faces and also to the inherent challenges in handling pose variations. In this work, we investigate how to learn a universal multi-view age estimator by harnessing 1) unlabeled web videos, 2) a publicly available labeled frontal face corpus, and 3) zero or more non-frontal faces with age labels. First, a large diverse human-involved video corpus is collected from online video sharing website. Then, multi-view face detection and tracking are performed to build a large set of frontal-vs-profile face bundles, each of which is from the same tracking sequence, and thus exhibiting the same age. These unlabeled face bundles constitute the so-called video context, and the parametric multi-view age estimator is trained by 1) enforcing the face-to-age relation for the partially labeled faces, 2) imposing the consistency of the predicted ages for the non-frontal and frontal faces within each face bundle, and 3) mutually constraining the multi-view age models with the spatial correspondence priors derived from the face bundles. Our multi-view age estimator performs well on a realistic evaluation dataset that contains faces under varying poses, and whose ground truth age was manually annotated.

In this article, beyond solo-activity analysis for single object, we study the more complicated pair-activity recognition problem by exploring the relationship between two active objects based on their trajectory clues obtained from video sensor. Our contributions are three-fold. First, we design two sets of features for representing the pair-activities encoded as length-variable trajectory pairs. One set characterizes the strength of causality between two trajectories, for example, the causality ratio and feedback ratio based on the Granger Causality Test (GCT), and another set describes the style of causality between two trajectories, for example, the sampled frequency responses of the digital filter with these two trajectories as the input and output discrete signals respectively. These features along with conventional velocity and position features of a trajectory-pair are essentially of multi-modalities, and may be greatly different in scales and importance. To make full use of them, we then develop a novel feature fusing procedure to learn the coefficients for weighting these features by maximizing the discriminating power measured by weighted correlation. Finally, we collected a pair-activity database of five popular categories, each of which consists of about 170 instances. The extensive experiments on this database validate the effectiveness of the designed features for pair-activity representation, and also demonstrate that the proposed feature fusing procedure significantly boosts the pair-activity classification accuracy.

In this paper, we present an automatic web image and video mining framework with the ultimate goal of building a universal human age estimator based on facial information, which is applicable to all ethnic groups and various image qualities. On one hand, a large (391 k) yet noisy human aging image database is collected from Flickr and Google Image using a set of human age-related text queries. Multiple human face detectors based on distinctive techniques are adopted for noise-prune face detection. For each image, the detected faces with high detection confidences constitute a bag of face instances. We further remove the outliers via principal component analysis (PCA), which results in a condensed image database with about 175 k face instances. A robust multi-instance regressor learning algorithm is then developed to learn the kernel regression-based human age estimator in the presence of bag label noises. On the other hand, about 10 k video clips are downloaded from YouTube. We extract tracked face sequences from these video clips. Although their age labels are unknown, the tracked faces within a sequence are naturally with identical ages. This age-consistence constraint for face pairs is used as an extra regularizer to enhance the robustness of the age estimator. The derived human age estimator is extensively evaluated on three benchmark human aging databases, and without taking any images from these benchmark databases as training samples, comparable age estimation accuracies with the state-of-the-art results are achieved.

In this paper, we propose an intelligent photography system, which automatically and professionally generates/recommends user-favorite photo(s) from a wide view or a continuous view sequence. This task is quite challenging given that the evaluation of photo quality is under-determined and usually subjective. Motivated by the recent prevalence of online media, we present a solution y mining the underlying knowledge and experience of the photographers from massively crawled professional photos (about 100,000 images, which are highly ranked by users) of those popular photo sharing websites, e.g. Flickr.com. Generally far contexts are critical in characterizing the composition rules for professional photos, and thus we present a method called omni-range context modeling to learn the patch/object spatial correlation distribution for the concurrent patch/object pair of arbitrary distance. The learned photo omni-range context priors then serve as rules to guide the composition of professional photos. When a wide view is fed into the system, these priors are utilized together with other cues (e.g., placements of faces at different poses, patch number, etc) to form a posterior probability formulation for professional sub-view finding. Moreover, this system can function as intelligent professionalview guider based on real-time view quality assessment and the embedded compass (for recording capture direction). Beyond the salient areas targeted by most existing view recommendation algorithms, the proposed system targets at professional photo composition. Qualitative experiments as well as comprehensive user studies well demonstrate the validity and efficiency of the proposed omnirange context learning method as well as the automatic view finding framework.

The aim of this paper is to address the problem of recognizing human group activities in surveillance videos. This task has great potentials in practice, however was rarely studied due to the lack of benchmark database and the difficulties caused by large intra-class variations. Our contributions are two-fold. Firstly, we propose to encode the group-activities with three types of localized causalities, namely self-causality, pair-causality, and group-causality, which characterize the local interaction/reasoning relations within, between, and among motion trajectories of different humans respectively. Each type of causality is expressed as a specific digital filter, whose frequency responses then constitute the feature representation space. Finally, each video clip of certain group activity is encoded as a bag of localized causalities/filters. We also collect a human group-activity video database, which involves six popular group activity categories with about 80 video clips for each in average, captured in five different sessions with varying numbers of participants. Extensive experiments on this database based on our proposed features and different classifiers show the promising results on this challenging task.

In this paper, we investigate how to incorporate spatial and/or temporal contextual information into classical histogram features with the aim of boosting visual classification performance. Firstly, we show that the stationary distribution derived from the normalized histogram-bin co-occurrence matrix characterizes the row sums of the original histogram-bin co-occurrence matrix. This underlying rationale of the histogram-bin co-occurrence features then motivates us to propose the concept of general contextualizing histogram process, which encodes the spatial and/or temporal contexts as local homogeneity distributions and produces the so called contextualized histograms by convoluting these local homogeneity distributions with the histogram-bin index images/videos. Finally, the third and even higher order contextualized histograms are instantiated for encoding more complicated and informative spatial and/or temporal contextual information into histograms. We evaluate these proposed methods on face recognition and group activity classification problems, and the results demonstrate that the contextualized histograms significantly boost the visual classification performance.

We investigate how to effectively incorporate spatial structure information into histogram features for boosting visual classification performance motivated by recently proposed Markov stationary features (MSF). First, we show that due to the symmetric property of the image occurrence modeling procedure, the stationary distribution derived from the normalized co-occurrence matrix has a trivial informative solution which only approximates the original histogram representation, i.e., does not encode proper spatial structure information. To eliminate this ambiguity, we propose in this work the so called directed Markov stationary features (DMSF) to encode spatial information into histogram features, and the asymmetric essence of the co-occurrence matrices in DMSF avoids the trivial informative solutions in MSF. Extensive experiments on face recognition show the significant performance improvement brought by our proposed DMSF.

In this paper, we present an automatic web image mining system towards building a universal human age estimator based on facial information, which is applicable to all ethnic groups and various image qualities. First, a large (<391k) yet noisy human aging image dataset is crawled from the photo sharing website Flickr and Google image search engine based on a set of human age related text queries. Then, within each image, several human face detectors of different implementations are used for robust face detection, and all the detected faces with multiple responses are considered as the multiple instances of a bag (image). An outlier removal step with Principal Component Analysis further refines the image set to about 220k faces, and then a robust multi-instance regressor learning algorithm is proposed to learn the kernel-regression based human age estimator under the scenarios with possibly noisy bags. The proposed system has the following characteristics: 1) no manual human age labeling process is required, and the age information is automatically obtained from the age related queries, 2) the derived human age estimator is universal owing to the diversity and richness of Internet images and thus has good generalization capability, and 3) the age estimator learning process is robust to the noises existing in both Internet images and corresponding age labels. This automatically derived human age estimator is extensively evaluated on three popular benchmark human aging databases, and without taking any images from these benchmark databases as training samples, comparable age estimation accuracies with the state-of-the-art results are achieved.

In this paper, we present a novel feature extraction framework, called learning by propagability. The whole learning process is driven by the philosophy that the data labels and optimal feature representation can constitute a harmonic system, namely, the data labels are invariant with respect to the propagation on the similarity-graph constructed by the optimal feature representation. Based on this philosophy, a unified formulation for learning by propagability is proposed for both supervised and semi-supervised configurations. Specifically, this formulation offers the semi-supervised learning two characteristics: 1) unlike conventional semi-supervised learning algorithms which mostly include at least two parameters, this formulation is parameter-free; and 2) the formulation unifies the label propagation and optimal representation pursuing, and thus the label propagation is enhanced by benefiting from the graph constructed with the derived optimal representation instead of the original representation. Extensive experiments on UCI toy data, handwritten digit recognition, and face recognition all validate the effectiveness of our proposed learning framework compared with the state-of-the-art methods for feature extraction and semi-supervised learning.

In this paper, we present a hybrid framework for articulated 3-D human motion tracking from multiple synchronized cameras with potential uses in surveillance systems. Although the recovery of 3-D motion provides richer information for event understanding, existing methods based on either deterministic search or stochastic sampling lack robustness or efficiency. We therefore propose a hybrid sample-and-refine framework that combines both stochastic sampling and deterministic optimization to achieve a good compromise between efficiency and robustness. Similar motion patterns are used to learn a compact low-dimensional representation of the motion statistics. Sampling in a low-dimensional space is implemented during tracking, which reduces the number of particles drastically. We also incorporate a local optimization method based on simulated physical force/moment into our framework, which further improves the optimality of the tracking. Experimental results on several real human motion sequences show the accuracy and robustness of our method, which also has a higher sampling efficiency than most particle filtering-based methods.