Vision-Language-Action (VLA) models have demonstrated significant potential in complex scene understanding and action reasoning, leading to their increasing adoption in end-to-end autonomous driving systems. However, the long visual tokens of VLA models greatly increase computational costs. Current visual token pruning methods in Vision-Language Models (VLM) rely on either visual token similarity or visual-text attention, but both have shown poor performance in autonomous driving scenarios. Given that human drivers concentrate on relevant foreground areas while driving, we assert that retaining visual tokens containing this foreground information is essential for effective decision-making. Inspired by this, we propose FastDriveVLA, a novel reconstruction-based vision token pruning framework designed specifically for autonomous driving. FastDriveVLA includes a plug-and-play visual token pruner called ReconPruner, which prioritizes foreground information through MAE-style pixel reconstruction. A novel adversarial foreground-background reconstruction strategy is designed to train ReconPruner for the visual encoder of VLA models. Once trained, ReconPruner can be seamlessly applied to different VLA models with the same visual encoder without retraining. To train ReconPruner, we also introduce a large-scale dataset called nuScenes-FG, consisting of 241K image-mask pairs with annotated foreground regions. Our approach achieves state-of-the-art results on the nuScenes open-loop planning benchmark across different pruning ratios.

Video Large Language Models (Video-LLMs) have demonstrated significant potential in the areas of video captioning, search, and summarization. However, current Video-LLMs still face challenges with long real-world videos. Recent methods have introduced a retrieval mechanism that retrieves query-relevant KV caches for question answering, enhancing the efficiency and accuracy of long real-world videos. However, the compression and retrieval of KV caches are still not fully explored. In this paper, we propose StreamKV, a training-free framework that seamlessly equips Video-LLMs with advanced KV cache retrieval and compression. Compared to previous methods that used uniform partitioning, StreamKV dynamically partitions video streams into semantic segments, which better preserves semantic information. For KV cache retrieval, StreamKV calculates a summary vector for each segment to retain segment-level information essential for retrieval. For KV cache compression, StreamKV introduces a guidance prompt designed to capture the key semantic elements within each segment, ensuring only the most informative KV caches are retained for answering questions. Moreover, StreamKV unifies KV cache retrieval and compression within a single module, performing both in a layer-adaptive manner, thereby further improving the effectiveness of streaming video question answering. Extensive experiments on StreamingVQA benchmarks demonstrate that StreamKV significantly outperforms existing Online Video-LLMs, achieving superior accuracy while substantially improving both memory efficiency and computational latency.

Data scarcity continues to be a critical bottleneck in the field of robotic manipulation, limiting the ability to train robust and generalizable models. While diffusion models provide a promising approach to synthesizing realistic robotic manipulation videos, their effectiveness hinges on the availability of precise and reasonable control instructions. Current methods primarily rely on 2D trajectories as instruction prompts, which inherently face issues with 3D spatial ambiguity. In this work, we present a novel framework named ManipDreamer3Dfor generating plausible 3D-aware robotic manipulation videos from the input image and the text instruction. Our method combines 3D trajectory planning with a reconstructed 3D occupancy map created from a third-person perspective, along with a novel trajectory-to-video diffusion model. Specifically, ManipDreamer3D first reconstructs the 3D occupancy representation from the input image and then computes an optimized 3D end-effector trajectory, minimizing path length, avoiding collisions and retiming. Next, we employ a latent editing technique to create video sequences from the initial image latent, text instruction and the optimized 3D trajectory. This process conditions our specially trained trajectory-to-video diffusion model to produce robotic pick-and-place videos. Our method significantly reduces human intervention requirements by autonomously planing plausible 3D trajectories. Experimental results demonstrate its superior visual quality and precision.

Video Large Language Models (VLLMs) excel in video understanding, but their excessive visual tokens pose a significant computational challenge for real-world applications. Current methods aim to enhance inference efficiency by visual token pruning. However, they do not consider the dynamic characteristics and temporal dependencies of video frames, as they perceive video understanding as a multi-frame task. To address these challenges, we propose MMG-Vid, a novel training-free visual token pruning framework that removes redundancy by Maximizing Marginal Gains at both segment-level and token-level. Specifically, we first divide the video into segments based on frame similarity, and then dynamically allocate the token budget for each segment to maximize the marginal gain of each segment. Subsequently, we propose a temporal-guided DPC algorithm that jointly models inter-frame uniqueness and intra-frame diversity, thereby maximizing the marginal gain of each token. By combining both stages, MMG-Vid can maximize the utilization of the limited token budget, significantly improving efficiency while maintaining strong performance. Extensive experiments demonstrate that MMG-Vid can maintain over 99.5% of the original performance, while effectively reducing 75% visual tokens and accelerating the prefilling stage by 3.9x on LLaVA-OneVision-7B.

While 3D Gaussian Splatting has shown promising results in street scene reconstruction, existing methods require massive numbers of Gaussian primitives to capture fine details, leading to prohibitive storage costs and slow rendering speeds. We observe that dynamic objects (e.g., vehicles and pedestrians) demand high-fidelity representations to maintain temporal consistency, while static background regions often contain substantial redundancy. Motivated by this, we propose SparseStreet, a general compression framework specifically designed for street scenes. First, we introduce a node-based learnable pruning strategy that systematically removes low-contributing Gaussian primitives while preserving visually critical regions. Second, after the scene representation stabilizes, we apply background compression, further reducing redundancy in static regions. Our method effectively preserves the geometry and appearance of dynamic objects while significantly reducing the total number of Gaussian primitives. Extensive experiments on the Waymo and nuScenes demonstrate that SparseStreet achieves up to 80% compression ratio with minimal quality degradation, enabling resource-efficient, high-fidelity dynamic scene reconstruction. Project website: https://sparsestreet.github.io/.

Thermography is especially valuable for the military and other users of surveillance cameras. Some recent methods based on Neural Radiance Fields (NeRF) have been proposed to reconstruct thermal scenes in 3D from a set of thermal and RGB images. However, unlike NeRF, 3D Gaussian splatting (3DGS) prevails due to its rapid training and real-time rendering. In this work, we propose ThermalGaussian, the first thermal 3DGS approach capable of rendering high-quality images in RGB and thermal modalities. We first calibrate the RGB camera and the thermal camera to ensure that both modalities are accurately aligned. Subsequently, we use the registered images to learn the multimodal 3D Gaussians. To prevent the overfitting of any single modality, we introduce several multimodal regularization constraints. We also develop smoothing constraints tailored to the physical characteristics of the thermal modality. Besides, we contribute a real-world dataset named RGBT-Scenes, captured by a handheld thermal-infrared camera, facilitating future research on thermal scene reconstruction. Based on ThermalGaussian, we further introduce ThermalGaussian++ to improve the alignment and resolution of ThermalGaussian. To improve multimodal alignment, we design a multimodal pose optimization module. This module enables direct processing of non-aligned multimodal image pairs, reducing the need for professional calibration before each use. To improve thermal resolution, we also propose a multimodal joint super-resolution reconstruction module, which enhances the quality of low-resolution thermal fields. Additionally, we contribute a new dataset: RGBT-Scenes++, which offers higher-resolution thermal images. We conduct comprehensive experiments demonstrating that ThermalGaussian++ achieves photorealistic thermal rendering and improves RGB rendering quality. It significantly enhances both alignment and resolution, enabling better practical deployment. In addition, our multimodal regularization constraints reduce the model's storage requirements. The code and datasets will be released.

Current compressed video super-resolution methods have achieved promising performance, but they often assume that an input video is compressed under low-delay configurations. However, under random access configurations, those methods might struggle to leverage the metadata effectively due to the large variations of metadata in different compression configurations. In this work, we propose a Compression-Oriented Video Super-Resolution (COVSR) method that can address video super-resolution for both low-delay and random-access configurations. Specifically, we first introduce an efficient compression-aware propagation (ECAP) module that dynamically adjusts propagation routes in accordance with the compression configurations. Since existing methods require reconstructing frames in a frame-by-frame manner, it is difficult to achieve efficient parallelization. However, we find that by slightly relaxing sequential dependencies, our ECAP can significantly improve inference speed. Furthermore, existing methods typically perform alignment between adjacent frames or adjacent features. However, since ECAP may propagate features along non-adjacent reference routes, it introduces new challenges for accurate cross-frame feature alignment. In response, we propose a metadata-driven alignment (MDA) module that refines cross-frame motion vectors into dense, feature-level flow offsets, enabling precise alignment across temporally distant features. Extensive experimental results demonstrate that our COVSR not only achieves efficient and superior super-resolution performance but also is generalizable to various compression configurations. Our code will be available at https://covsr.github.io.

Rendering photorealistic head avatars from arbitrary viewpoints is crucial for various applications like virtual reality. Although previous methods based on Neural Radiance Fields (NeRF) can achieve impressive results, they lack fidelity and efficiency. Recent methods using 3D Gaussian Splatting (3DGS) have improved rendering quality and real-time performance but still require significant storage overhead. In this paper, we introduce a method called GraphAvatar that utilizes Graph Neural Networks (GNN) to generate 3D Gaussians for the head avatar. Specifically, GraphAvatar trains a geometric GNN and an appearance GNN to generate the attributes of the 3D Gaussians from the tracked mesh. Therefore, our method can store the GNN models instead of the 3D Gaussians, significantly reducing the storage overhead to just 10MB. To reduce the impact of face-tracking errors, we also present a novel graph-guided optimization module to refine face-tracking parameters during training. Finally, we introduce a 3D-aware enhancer for post-processing to enhance the rendering quality. We conduct comprehensive experiments to demonstrate the advantages of GraphAvatar, surpassing existing methods in visual fidelity and storage consumption. The ablation study sheds light on the trade-offs between rendering quality and model size.

Visual encoders are fundamental components in vision-language models (VLMs), each showcasing unique strengths derived from various pre-trained visual foundation models. To leverage the various capabilities of these encoders, recent studies incorporate multiple encoders within a single VLM, leading to a considerable increase in computational cost. In this paper, we present Mixture-of-Visual-Encoder Knowledge Distillation (MoVEKD), a novel framework that distills the unique proficiencies of multiple vision encoders into a single, efficient encoder model. Specifically, to mitigate conflicts and retain the unique characteristics of each teacher encoder, we employ low-rank adaptation (LoRA) and mixture-of-experts (MoEs) to selectively activate specialized knowledge based on input features, enhancing both adaptability and efficiency. To regularize the KD process and enhance performance, we propose an attention-based distillation strategy that adaptively weighs the different encoders and emphasizes valuable visual tokens, reducing the burden of replicating comprehensive but distinct features from multiple teachers. Comprehensive experiments on popular VLMs, such as LLaVA and LLaVA-NeXT, validate the effectiveness of our method. Our code is available at: https://github.com/hey-cjj/MoVE-KD.

Gaussian Splatting (GS) methods, including 3DGS and 2DGS, have demonstrated exceptional performance in real-time novel view synthesis (NVS), emerging as a transformative technology in the fields of explicit rendering and computer graphics. However, GS-based methods still face challenges in rendering high-quality image details. Even when using high-quality training frameworks, their outputs often exhibit severe rendering artifacts, such as noise and blurriness. A reasonable approach is to perform post-processing to restore clear details. In this paper, we propose GaussianEnhancer, a general network-agnostic post-processor that employs a degradation-driven view blending method to improve the rendering quality of GS models while preserving the original network’s performance. Specifically, we design a degradation modeling method tailored to the GS-style and construct a large-scale training dataset to effectively simulate the native rendering artifacts of GS, enabling efficient training. In addition, we introduce a spatial information fusion framework, consisting of view fusion and depth modulation modules, which can blend highly correlated high-quality training images and leverage the depth information of the target image to complete the rendering details. Through our GaussianEnhancer, we are able to effectively eliminate the rendering artifacts of GS models and generate highly realistic synthetic views.

3D cars are commonly used in self-driving systems, virtual/augmented reality, and games. However, existing 3D car datasets are either synthetic or low-quality, limiting their applications in practical scenarios and presenting a significant gap toward high-quality real-world 3D car datasets. In this paper, we propose the first large-scale 3D real car dataset, termed 3DRealCar, offering three distinctive features. (1) \textbf{High-Volume}: 2,500 cars are meticulously scanned by smartphones, obtaining car images and point clouds with real-world dimensions; (2) \textbf{High-Quality}: Each car is captured in an average of 200 dense, high-resolution 360-degree RGB-D views, enabling high-fidelity 3D reconstruction; (3) \textbf{High-Diversity}: The dataset contains various cars from over 100 brands, collected under three distinct lighting conditions, including reflective, standard, and dark. Additionally, we offer detailed car parsing maps for each instance to promote research in car parsing tasks. Moreover, we remove background point clouds and standardize the car orientation to a unified axis for the reconstruction only on cars and controllable rendering without background. We benchmark 3D reconstruction results with state-of-the-art methods across different lighting conditions in 3DRealCar. Extensive experiments demonstrate that the standard lighting condition part of 3DRealCar can be used to produce a large number of high-quality 3D cars, improving various 2D and 3D tasks related to cars. Notably, our dataset brings insight into the fact that recent 3D reconstruction methods face challenges in reconstructing high-quality 3D cars under reflective and dark lighting conditions. \textcolor{red}{\href{https://xiaobiaodu.github.io/3drealcar/}{Our dataset is here.}}

Gaze estimation encounters generalization challenges when dealing with out-of-distribution data. To address this problem, recent methods use neural radiance fields (NeRF) to generate augmented data. However, existing methods based on NeRF are computationally expensive and lack facial details. 3D Gaussian Splatting (3DGS) has become the prevailing representation of neural fields. While 3DGS has been extensively examined in head avatars, it faces challenges with accurate gaze control and generalization across different subjects. In this work, we propose GazeGaussian, the first high-fidelity gaze redirection method that uses a two-stream 3DGS model to represent the face and eye regions separately. Leveraging the unstructured nature of 3DGS, we develop a novel representation of the eye for rigid eye rotation based on the target gaze direction. To enable synthesis generalization across various subjects, we integrate an expression-guided module to inject subject-specific information into the neural renderer. Comprehensive experiments show that GazeGaussian outperforms existing methods in rendering speed, gaze redirection accuracy, and facial synthesis across multiple datasets. The code is available at: https://ucwxb.github.io/GazeGaussian.

Photorealistic reconstruction of street scenes is essential for developing real-world simulators in autonomous driving. While recent methods based on 3D/4D Gaussian Splatting (GS) have demonstrated promising results, they still encounter challenges in complex street scenes due to the unpredictable motion of dynamic objects. Current methods typically decompose street scenes into static and dynamic objects, learning the Gaussians in either a supervised manner (e.g., w/ 3D bounding-box) or a self-supervised manner (e.g., w/o 3D bounding-box). However, these approaches do not effectively model the motions of dynamic objects (e.g., the motion speed of pedestrians is clearly different from that of vehicles), resulting in suboptimal scene decomposition. To address this, we propose Explicit Motion Decomposition (EMD), which models the motions of dynamic objects by introducing learnable motion embeddings to the Gaussians, enhancing the decomposition in street scenes. The proposed plug-and-play EMD module compensates for the lack of motion modeling in self-supervised street Gaussian splatting methods. We also introduce tailored training strategies to extend EMD to supervised approaches. Comprehensive experiments demonstrate the effectiveness of our method, achieving state-of-the-art novel view synthesis performance in self-supervised settings. The code is available at: https://qingpowuwu.github.io/emd.

Thermography is especially valuable for the military and other users of surveillance cameras. Some recent methods based on Neural Radiance Fields (NeRF) are proposed to reconstruct the thermal scenes in 3D from a set of thermal and RGB images. However, unlike NeRF, 3D Gaussian splatting (3DGS) prevails due to its rapid training and real-time rendering. In this work, we propose ThermalGaussian, the first thermal 3DGS approach capable of rendering high-quality images in RGB and thermal modalities. We first calibrate the RGB camera and the thermal camera to ensure that both modalities are accurately aligned. Subsequently, we use the registered images to learn the multimodal 3D Gaussians. To prevent the overfitting of any single modality, we introduce several multimodal regularization constraints. We also develop smoothing constraints tailored to the physical characteristics of the thermal modality. Besides, we contribute a real-world dataset named RGBT-Scenes, captured by a hand-hold thermal-infrared camera, facilitating future research on thermal scene reconstruction. We conduct comprehensive experiments to show that ThermalGaussian achieves photorealistic rendering of thermal images and improves the rendering quality of RGB images. With the proposed multimodal regularization constraints, we also reduced the model's storage cost by 90\%. Our project page is at https://thermalgaussian.github.io/.

Real-time speech-driven 3D facial animation has been attractive in academia and industry. Traditional methods mainly focus on learning a deterministic mapping from speech to animation. Recent approaches start to consider the nondeterministic fact of speech-driven 3D face animation and employ the diffusion model for the task. Existing diffusion-based methods can improve the diversity of facial animation. However, personalized speaking styles conveying accurate lip language is still lacking, besides, efficiency and compactness still need to be improved. In this work, we propose DiffusionTalker to address the above limitations via personalizer-guided distillation. In terms of personalization, we introduce a contrastive personalizer that learns identity and emotion embeddings to capture speaking styles from audio. We further propose a personalizer enhancer during distillation to enhance the influence of embeddings on facial animation. For efficiency, we use iterative distillation to reduce the steps required for animation generation and achieve more than 8x speedup in inference. To achieve compactness, we distill the large teacher model into a smaller student model, reducing our model’s storage by 86.4% while minimizing performance loss. After distillation, users can derive their identity and emotion embeddings from audio to quickly create personal-ized animations that reflect specific speaking styles. Extensive experiments are conducted to demonstrate that our method outperforms state-of-the-art methods. The code is released at: https://github.com/ChenVoid/DiffusionTalker.

3D semantic occupancy prediction is a crucial task in visual perception, as it requires the simultaneous comprehension of both scene geometry and semantics. It plays a crucial role in understanding 3D scenes and has great potential for various applications, such as robotic vision perception and autonomous driving. Many existing works utilize planar-based representations such as Bird's Eye View (BEV) and Tri-Perspective View (TPV). These representations aim to simplify the complexity of 3D scenes while preserving essential object information, thereby facilitating efficient scene representation. However, in dense indoor environments with prevalent occlusions, directly applying these planar-based methods often leads to difficulties in capturing global semantic occupancy, ultimately degrading model performance. In this paper, we present a new vertical slice representation that divides the scene along the vertical axis and projects spatial point features onto the nearest pair of parallel planes. To utilize these slice features, we propose SliceOcc, an RGB camera-based model specifically tailored for indoor 3D semantic occupancy prediction. SliceOcc utilizes pairs of slice queries and cross-attention mechanisms to extract planar features from input images. These local planar features are then fused to form a global scene representation, which is employed for indoor occupancy prediction. Experimental results on the EmbodiedScan dataset demonstrate that SliceOcc achieves a mIoU of 15.45 % across 81 indoor categories, setting a new state-of-the-art performance among RGB camera-based models for indoor 3D semantic occupancy prediction.

Neural fields are now the central focus of research in 3D vision and computer graphics. Existing methods mainly focus on various scene representations, such as neural points and 3D Gaussians. However, few works have studied the rendering process to enhance the neural fields. In this work, we propose a plug-in method named K-Buffers that leverages multiple buffers to improve the rendering performance. Our method first renders K buffers from scene representations and constructs K pixel-wise feature maps. Then, We introduce a K-Feature Fusion Network (KFN) to merge the K pixel-wise feature maps. Finally, we adopt a feature decoder to generate the rendering image. We also introduce an acceleration strategy to improve rendering speed and quality. We apply our method to well-known radiance field baselines, including neural point fields and 3D Gaussian Splatting (3DGS). Extensive experiments demonstrate that our method effectively enhances the rendering performance of neural point fields and 3DGS.

Reconstructing high-fidelity 3D head avatars is crucial in various applications such as virtual reality. The pioneering methods reconstruct realistic head avatars with Neural Radiance Fields (NeRF), which have been limited by training and rendering speed. Recent methods based on 3D Gaussian Splatting (3DGS) significantly improve the efficiency of training and rendering. However, the surface inconsistency of 3DGS results in subpar geometric accuracy; later, 2DGS uses 2D surfels to enhance geometric accuracy at the expense of rendering fidelity. To leverage the benefits of both 2DGS and 3DGS, we propose a novel method named MixedGaussianAvatar for realistically and geometrically accurate head avatar reconstruction. Our main idea is to utilize 2D Gaussians to reconstruct the surface of the 3D head, ensuring geometric accuracy. We attach the 2D Gaussians to the triangular mesh of the FLAME model and connect additional 3D Gaussians to those 2D Gaussians where the rendering quality of 2DGS is inadequate, creating a mixed 2D-3D Gaussian representation. These 2D-3D Gaussians can then be animated using FLAME parameters. We further introduce a progressive training strategy that first trains the 2D Gaussians and then fine-tunes the mixed 2D-3D Gaussians. We use a unified mixed Gaussian representation to integrate the two modalities of 2D image and 3D mesh. Furthermore, the comprehensive experiments demonstrate the superiority of MixedGaussianAvatar. The code will be released.

Sparse-view 3D reconstruction is a fundamental yet challenging task in practical 3D reconstruction applications. Recently, many methods based on 3D Gaussian Splatting (3DGS) have been proposed to address sparse-view 3D reconstruction. Although these methods have made considerable advancements, they still show significant issues with overfitting. To reduce the overfitting, we introduce VGNC, a novel Validation-guided Gaussian Number Control approach based on generative novel view synthesis (NVS) models. To the best of our knowledge, this is the first attempt to alleviate the overfitting issue of sparse-view 3DGS with generative validation images. Specifically, we first introduce a validation image generation method based on a generative NVS model. We then propose a Gaussian number control strategy that utilizes generated validation images to determine optimal Gaussian numbers, thereby reducing the issue of overfitting. We conducted detailed experiments on various sparse-view 3DGS baselines and datasets to evaluate the effectiveness of VGNC. Extensive experiments show that our approach not only reduces overfitting but also improves rendering quality on the test set while decreasing the number of Gaussians. This reduction lowers storage demands and accelerates both training and rendering. Our code is available at: https://github.com/LinLif1869/VGNC.

Online 3D occupancy prediction provides a comprehensive spatial understanding of embodied environments. While the innovative EmbodiedOcc framework utilizes 3D semantic Gaussians for progressive indoor occupancy prediction, it overlooks the geometric characteristics of indoor environments, which are primarily characterized by planar structures. This paper introduces EmbodiedOcc++, enhancing the original framework with two key innovations: a Geometry-guided Refinement Module (GRM) that constrains Gaussian updates through plane regularization, along with a Semantic-aware Uncertainty Sampler (SUS) that enables more effective updates in overlapping regions between consecutive frames. GRM regularizes the position update to align with surface normals. It determines the adaptive regularization weight using curvature-based and depth-based constraints, allowing semantic Gaussians to align accurately with planar surfaces while adapting in complex regions. To effectively improve geometric consistency from different views, SUS adaptively selects proper Gaussians to update. Comprehensive experiments on the EmbodiedOcc-ScanNet benchmark demonstrate that EmbodiedOcc++ achieves state-of-the-art performance across different settings. Our method demonstrates improved edge accuracy and retains more geometric details while ensuring computational efficiency, which is essential for online embodied perception. The code will be released at: https://github.com/PKUHaoWang/EmbodiedOcc2.

In recent years, foundation models for monocular depth estimation have received increasing attention. Current methods mainly address typical daylight conditions, but their effectiveness notably decreases in low-light environments. There is a lack of robust foundational models for monocular depth estimation specifically designed for low-light scenarios. This largely stems from the absence of large-scale, high-quality paired depth datasets for low-light conditions and the effective parameter-efficient fine-tuning (PEFT) strategy. To address these challenges, we propose DepthDark, a robust foundation model for low-light monocular depth estimation. We first introduce a flare-simulation module and a noise-simulation module to accurately simulate the imaging process under nighttime conditions, producing high-quality paired depth datasets for low-light conditions. Additionally, we present an effective low-light PEFT strategy that utilizes illumination guidance and multiscale feature fusion to enhance the model's capability in low-light environments. Our method achieves state-of-the-art depth estimation performance on the challenging nuScenes-Night and RobotCar-Night datasets, validating its effectiveness using limited training data and computing resources.

Personalized models have demonstrated remarkable success in understanding and generating concepts provided by users. However, existing methods use separate concept tokens for understanding and generation, treating these tasks in isolation. This may result in limitations for generating images with complex prompts. For example, given the concept $\langle bo\rangle$, generating "$\langle bo\rangle$ wearing its hat" without additional textual descriptions of its hat. We call this kind of generation \textit{\textbf{personalized attribute-reasoning generation}}. To address the limitation, we present UniCTokens, a novel framework that effectively integrates personalized information into a unified vision language model (VLM) for understanding and generation. UniCTokens trains a set of unified concept tokens to leverage complementary semantics, boosting two personalized tasks. Moreover, we propose a progressive training strategy with three stages: understanding warm-up, bootstrapping generation from understanding, and deepening understanding from generation to enhance mutual benefits between both tasks. To quantitatively evaluate the unified VLM personalization, we present UnifyBench, the first benchmark for assessing concept understanding, concept generation, and attribute-reasoning generation. Experimental results on UnifyBench indicate that UniCTokens shows competitive performance compared to leading methods in concept understanding, concept generation, and achieving state-of-the-art results in personalized attribute-reasoning generation. Our research demonstrates that enhanced understanding improves generation, and the generation process can yield valuable insights into understanding. Our code and dataset will be released at: \href{https://github.com/arctanxarc/UniCTokens}{https://github.com/arctanxarc/UniCTokens}.

In multimodal large language models (MLLMs), the length of input visual tokens is often significantly greater than that of their textual counterparts, leading to a high inference cost. Many works aim to address this issue by removing redundant visual tokens. However, current approaches either rely on attention-based pruning, which retains numerous duplicate tokens, or use similarity-based pruning, overlooking the instruction relevance, consequently causing suboptimal performance. In this paper, we go beyond attention or similarity by proposing a novel visual token pruning method named **CDPruner**, which maximizes the conditional diversity of retained tokens. We first define the conditional similarity between visual tokens conditioned on the instruction, and then reformulate the token pruning problem with determinantal point process (DPP) to maximize the conditional diversity of the selected subset. The proposed CDPruner is training-free and model-agnostic, allowing easy application to various MLLMs. Extensive experiments across diverse MLLMs show that CDPruner establishes new state-of-the-art on various vision-language benchmarks. By maximizing conditional diversity through DPP, the selected subset better represents the input images while closely adhering to user instructions, thereby preserving strong performance even with high reduction ratios. When applied to LLaVA, CDPruner reduces FLOPs by **95\%** and CUDA latency by **78\%**, while maintaining **94\%** of the original accuracy. Our code is available at https://github.com/Theia-4869/CDPruner.

Previous methods utilize the Neural Radiance Field (NeRF) for panoptic lifting, while their training and rendering speed are unsatisfactory. In contrast, 3D Gaussian Splatting (3DGS) has emerged as a prominent technique due to its rapid training and rendering speed. However, unlike NeRF, the conventional 3DGS may not satisfy the basic smoothness assumption as it does not rely on any parameterized structures to render (e.g., MLPs). Consequently, the conventional 3DGS is, in nature, more susceptible to noisy 2D mask supervision. In this paper, we propose a new method called PLGS that enables 3DGS to generate consistent panoptic segmentation masks from noisy 2D segmentation masks while maintaining superior efficiency compared to NeRF-based methods. Specifically, we build a panoptic-aware structured 3D Gaussian model to introduce smoothness and design effective noise reduction strategies. For the semantic field, instead of initialization with structure from motion, we construct reliable semantic anchor points to initialize the 3D Gaussians. We then use these anchor points as smooth regularization during training. Additionally, we present a self-training approach using pseudo labels generated by merging the rendered masks with the noisy masks to enhance the robustness of PLGS. For the instance field, we project the 2D instance masks into 3D space and match them with oriented bounding boxes to generate cross-view consistent instance masks for supervision. Experiments on various benchmarks demonstrate that our method outperforms previous state-of-the-art methods in terms of both segmentation quality and speed.

Neural 3D reconstruction from multi-view images has recently attracted increasing attention from the community. Existing methods normally learn a neural field for the whole scene, while it is still under-explored how to reconstruct a target object indicated by users. Considering the Segment Anything Model (SAM) has shown effectiveness in segmenting any 2D images, in this paper, we propose NTO3D, a novel high-quality Neural Target Object 3D (NTO3D) reconstruction method, which leverages the benefits of both neural field and SAM. We first propose a novel strategy to lift the multi-view 2D segmentation masks of SAM into a unified 3D occupancy field. The 3D occupancy field is then projected into 2D space and generates the new prompts for SAM. This process is iterative until convergence to sepa-rate the target object from the scene. After this, we then lift the 2D features of the SAM encoder into a 3D feature field in order to improve the reconstruction quality of the target object. NTO3D lifts the 2D masks and features of SAM into the 3D neural field for high-quality neural target object 3D reconstruction. We conduct detailed experiments on several benchmark datasets to demonstrate the advantages of our method. The code will be available at: https://github.com/ucwxb/NTO3D.

Since real-world machine systems are running in non-stationary environments, Continual Test-Time Adaptation (CTTA) task is proposed to adapt the pre-trained model to continually changing target domains. Recently, existing methods mainly focus on model-based adaptation, which aims to leverage a self-training manner to extract the target domain knowledge. However, pseudo labels can be noisy and the updated model parameters are unreliable under dynamic data distributions, leading to error accumulation and catastrophic forgetting in the continual adaptation process. To tackle these challenges and maintain the model plasticity, we design a Visual Domain Adapter (ViDA) for CTTA, explicitly handling both domain-specific and domain-shared knowledge. Specifically, we first comprehensively explore the different domain representations of the adapters with trainable high-rank or low-rank embedding spaces. Then we inject ViDAs into the pre-trained model, which leverages high-rank and low-rank features to adapt the current domain distribution and maintain the continual domain-shared knowledge, respectively. To exploit the low-rank and high-rank ViDAs more effectively, we further propose a Homeostatic Knowledge Allotment (HKA) strategy, which adaptively combines different knowledge from each ViDA. Extensive experiments conducted on four widely used benchmarks demonstrate that our proposed method achieves state-of-the-art performance in both classification and segmentation CTTA tasks. Note that, our method can be regarded as a novel transfer paradigm for large-scale models, delivering promising results in adaptation to continually changing distributions.

Vision-centric bird-eye-view (BEV) perception has shown promising potential in autonomous driving. Recent works mainly focus on improving efficiency or accuracy but neglect the challenges when facing environment changing, resulting in severe degradation of transfer performance. For BEV perception, we figure out the significant domain gaps existing in typical real-world cross-domain scenarios and comprehensively solve the Domain Adaption (DA) problem for multi-view 3D object detection. Since BEV perception approaches are complicated and contain several components, the domain shift accumulation on multiple geometric spaces (i.e., 2D, 3D Voxel, BEV) makes BEV DA even challenging. In this paper, we propose a Multi-space Alignment Teacher-Student (MATS) framework to ease the domain shift accumulation, which consists of a Depth-Aware Teacher (DAT) and a Geometric-space Aligned Student (GAS) model. DAT tactfully combines target lidar and reliable depth prediction to construct depth-aware information, extracting target domain-specific knowledge in Voxel and BEV feature spaces. It then transfers the sufficient domain knowledge of multiple spaces to the student model. In order to jointly alleviate the domain shift, GAS projects multi-geometric space features to a shared geometric embedding space and decreases data distribution distance between two domains. To verify the effectiveness of our method, we conduct BEV 3D object detection experiments on three cross-domain scenarios and achieve state-of-the-art performance. Code: https://github.com/liujiaming1996/BEVUDA.

Neuromorphic spike data, an upcoming modality with high temporal resolution, has shown promising potential in autonomous driving by mitigating the challenges posed by high-velocity motion blur. However, training the spike depth estimation network holds significant challenges in two aspects: sparse spatial information for pixel-wise tasks and difficulties in achieving paired depth labels for temporally intensive spike streams. Therefore, we introduce open-source RGB data to support spike depth estimation, leveraging its annotations and spatial information. The inherent differences in modalities and data distribution make it challenging to directly apply transfer learning from open-source RGB to target spike data. To this end, we propose a cross-modality cross-domain (BiCross) framework to realize unsupervised spike depth estimation by introducing simulated mediate source spike data. Specifically, we design a Coarse-to-Fine Knowledge Distillation (CFKD) approach to facilitate comprehensive cross-modality knowledge transfer while preserving the unique strengths of both modalities, utilizing a spike-oriented uncertainty scheme. Then, we propose a Self-Correcting Teacher-Student (SCTS) mechanism to screen out reliable pixel-wise pseudo labels and ease the domain shift of the student model, which avoids error accumulation in target spike data. To verify the effectiveness of BiCross, we conduct extensive experiments on four scenarios, including Synthetic to Real, Extreme Weather, Scene Changing, and Real Spike. Our method achieves state-of-the-art (SOTA) performances, compared with RGB-oriented unsupervised depth estimation methods. Code and dataset: https://github.com/Theia-4869/BiCross.

In recent years, neural field-based methods for synthesizing novel views have gained popularity due to their exceptional rendering quality and fast training speed. However, the computational cost of volumetric rendering has significantly increased with the advancement of camera technology and the subsequent rise in average camera resolution. Despite extensive efforts to accelerate the training process, the training duration remains unacceptable for high-resolution inputs. Therefore, it's crucial to develop efficient sampling methods to optimize the learning process of neural fields from large inputs. In this paper, we present a new technique called Superpixel-based Efficient Sampling (SES) to improve the learning efficiency of neural fields. Our approach optimizes pixel-level ray sampling by segmenting the error map into multiple superpixels and dynamically updating their errors during training to increase ray sampling in superpixel areas with higher rendering errors. Compared with other methods, our approach leverages the flexibility of superpixels, effectively reducing redundant sampling while considering local information. Our method not only speeds up the learning process but also enhances the rendering quality learned from large inputs. We conduct extensive experiments to evaluate the effectiveness of our method across several baselines and datasets. The code will be released.

Bird'View (BEV) 3D Object Detection is a crucial multi-view technique for autonomous driving systems. Recently, plenty of works are proposed, following a similar paradigm consisting of three essential components, i.e., camera feature extraction, BEV feature construction, and task heads. Among the three components, BEV feature construction is BEV-specific compared with 2D tasks. Existing methods aggregate the multi-view camera features to the flattened grid in order to construct the BEV feature. However, flattening the BEV space along the height dimension fails to emphasize the informative features of different heights. For example, the barrier is located at a low height while the truck is located at a high height. In this paper, we propose a novel method named BEV Slice Attention Network (BEV-SAN) for exploiting the intrinsic characteristics of different heights. Instead of flattening the BEV space, we first sample along the height dimension to build the global and local BEV slices. Then, the features of BEV slices are aggregated from the camera features and merged by the attention mechanism. Finally, we fuse the merged local and global BEV features by a transformer to generate the final feature map for task heads. The purpose of local BEV slices is to emphasize informative heights. In order to find them, we further propose a LiDAR-guided sampling strategy to leverage the statistical distribution of LiDAR to determine the heights of local slices. Compared with uniform sampling, LiDAR-guided sampling can determine more informative heights. We conduct detailed experiments to demonstrate the effectiveness of BEV-SAN. Code will be released.

With the development of high-definition display devices, the practical scenario of Super-Resolution (SR) usually needs to super-resolve large input like 2K to higher resolution (4K/8K). To reduce the computational and memory cost, current methods first split the large input into local patches and then merge the SR patches into the output. These methods adaptively allocate a subnet for each patch. Quantization is a very important technique for network acceleration and has been used to design the subnets. Current methods train an MLP bit selector to determine the propoer bit for each layer. However, they uniformly sample subnets for training, making simple subnets overfitted and complicated subnets underfitted. Therefore, the trained bit selector fails to determine the optimal bit. Apart from this, the introduced bit selector brings additional cost to each layer of the$SR$network. In this paper, we propose a novel method named Content-Aware Bit Mapping (CABM), which can remove the bit selector without any performance loss. CABM also learns a bit selector for each layer during training. After training, we analyze the relation between the edge information of an input patch and the bit of each layer. We observe that the edge information can be an effective metric for the selected bit. Therefore, we design a strategy to build an Edge-to-Bit lookup table that maps the edge score of a patch to the bit of each layer during inference. The bit configuration of SR network can be determined by the lookup tables of all layers. Our strategy can find better bit configuration, resulting in more efficient mixed precision networks. We conduct detailed experiments to demonstrate the generalization ability of our method. The code will be released.

Super-Resolution (SR) has gained increasing research attention over the past few years. With the development of Deep Neural Networks (DNNs), many super-resolution methods based on DNNs have been proposed. Although most of these methods are aimed at ordinary frames, there are few works on super-resolution of omnidirectional frames. In these works, omnidirectional frames are projected from the 3D sphere to a 2D plane by Equi-Rectangular Projection (ERP). Although ERP has been widely used for projection, it has severe projection distortion near poles. Current DNN-based SR methods use 2D convolution modules, which is more suitable for the regular grid. In this paper, we find that different projection methods have great impact on the performance of DNNs. To study this problem, a comprehensive comparison of projections in omnidirectional super-resolution is conducted. We compare the SR results of different projection methods. Experimental results show that Equi-Angular cube map projection (EAC), which has minimal distortion, achieves the best result in terms of WS-PSNR compared with other projections. Code and data will be released.

Multi-view 3D detection based on BEV (bird-eye-view) has recently achieved significant improvements. However, the huge memory consumption of state-of-the-art models makes it hard to deploy them on vehicles, and the nontrivial latency will affect the real-time perception of streaming applications. Despite the wide application of quantization to lighten models, we show in our paper that directly applying quantization in BEV tasks will 1) make the training unstable, and 2) lead to intolerable performance degradation. To solve these issues, our method QD-BEV enables a novel view-guided distillation (VGD) objective, which can stabilize the quantization-aware training (QAT) while enhancing the model performance by leveraging both image features and BEV features. Our experiments show that QD-BEV achieves similar or even better accuracy than previous methods with significant efficiency gains. On the nuScenes datasets, the 4-bit weight and 6-bit activation quantized QD-BEV-Tiny model achieves 37.2% NDS with only 15.8 MB model size, outperforming BevFormer-Tiny by 1.8% with an 8× model compression. On the Small and Base variants, QD-BEV models also perform superbly and achieve 47.9% NDS (28.2 MB) and 50.9% NDS (32.9 MB), respectively.

Face retouching is a crucial technique for many consumer-level products. The goal of face retouching is to remove skin imperfections and preserve facial details simultaneously. However, it usually requires tedious manual work to achieve professional retouching effect. With the advent of Deep Neural Networks (DNNs), some methods were recently proposed to complete the task of face retouching automatically by using DNNs. They divide a portrait photo into local patches and train a DNN for face retouching. Although they can produce professional results automatically, there are still some limitations. Firstly, the network architecture fails to preserve sufficient facial details. Secondly, the facial semantic information is ignored when dividing a photo into some local patches. In this paper, we propose a novel method to solve these limitations. We first introduce the Masked Feature Fusion (MFF) module to a UNet, enabling the network to better preserve details in facial regions. Then, we exploit the semantic information by the Semantic-Aware Modulation (SAM) module, further boosting the retouching performance. Experiments on the recent public dataset Flickr-Faces-HQ-Retouched (FFHQR) demonstrate the effectiveness of our method. The code will be released.

Recently, content-aware methods have been utilized to reduce the bandwidth and improve the quality of Internet video delivery. Existing methods train corresponding content-aware super-resolution (SR) models for each video chunk on the server and stream low-resolution (LR) video chunks along with SR models to the client. Previous works introduce additional partial parameters to privatize the models of different video chunks. However, this still leads to the accumulation of parameters and even fails to modulate when the length of video increases, bringing extra delivery costs and performance degradation. In this paper, we introduce a novel Re-parameterization Content-aware Modulation (RepCaM) method to modulate all the video chunks with an end-to-end training strategy. Our method adopts extra parallel-cascade parameters during training to fit multiple chunks while removing the additional parameters through re-parameterization during inference. Therefore, RepCaM increases no extra model size compared with the original SR model. Moreover, in order to improve the training efficiency on servers, we propose an online Video Patch Sampling (VPS) method to speed up the training convergence. We conduct extensive experiments on VSD4K and newly collected dataset (VSD4K-2022), achieving state-of-the-art results in video restoration quality and delivery bandwidth compression. Code is available at: https://github.com/Neural-video-delivery/RepCaM-Pytorch-NOSSDAV2023.

Parametric face models, such as morphable and blendshape models, have shown great potential in face representation, reconstruction, and animation. However, all these models focus on large-scale facial geometry. Facial details such as wrinkles are not parameterized in these models, impeding accuracy and realism. In this article, we propose a method to learn a Semantically Disentangled Variational Autoencoder (SDVAE) to parameterize facial details and support independent detail manipulation as an extension of an off-the-shelf large-scale face model. Our method utilizes the non-linear capability of Deep Neural Networks for detail modeling, achieving better accuracy and greater representation power compared with linear models. In order to disentangle the semantic factors of identity, expression and age, we propose to eliminate the correlation between different factors in an adversarial manner. Therefore, wrinkle-level details of various identities, expressions, and ages can be generated and independently controlled by changing latent vectors of our SDVAE. We further leverage our model to reconstruct 3D faces via fitting to facial scans and images. Benefiting from our parametric model, we achieve accurate and robust reconstruction, and the reconstructed details can be easily animated and manipulated. We evaluate our method on practical applications, including scan fitting, image fitting, video tracking, model manipulation, and expression and age animation. Extensive experiments demonstrate that the proposed method can robustly model facial details and achieve better results than alternative methods.

Blendshape representations are widely used in facial animation. Consistent semantics must be maintained for all the blendshapes to build the blendshapes of one character. However, this is difficult for real characters because the face shape of the same semantics varies significantly across identities. Previous studies have handled this issue by asking users to perform a set of predefined expressions with specified semantics. We observe that facial emotions can be used to define semantics. Herein, we propose a real-time technique that directly updates blendshapes without predefined expressions. Its aim is to preserve semantics based on the emotion information extracted from an arbitrary facial motion sequence. In addition, we have designed corresponding algorithms to efficiently update blendshapes with large- and middle-scale face shapes and fine-scale facial details, such as wrinkles, in a real-time face tracking system. The experimental results indicate that using a commodity RGBD sensor, we can achieve real-time online blendshape updates with well-preserved semantics and user-specific facial features and details.

Internet video delivery has undergone a tremendous explosion of growth over the past few years. However, the quality of video delivery system greatly depends on the Internet bandwidth. Deep Neural Networks (DNNs) are utilized to improve the quality of video delivery recently. These methods divide a video into chunks, and stream LR video chunks and corresponding content-aware models to the client. The client runs the inference of models to super-resolve the LR chunks. Consequently, a large number of models are streamed in order to deliver a video. In this paper, we first carefully study the relation between models of different chunks, then we tactfully design a joint training framework along with the Content-aware Feature Modulation (CaFM) layer to compress these models for neural video delivery. With our method, each video chunk only requires less than 1% of original parameters to be streamed, achieving even better SR performance. We conduct extensive experiments across various SR backbones, video time length, and scaling factors to demonstrate the advantages of our method. Besides, our method can be also viewed as a new approach of video coding. Our primary experiments achieve better video quality compared with the commercial H.264 and H.265 standard under the same storage cost, showing the great potential of the proposed method. Code is available at: https://github.com/Neural-video-delivery/ CaFM-Pytorch-ICCV2021

Convolutional neural networks have been proven effective in a variety of image restoration tasks. Most state-of-the-art solutions, however, are trained using images with a single particular degradation level, and their performance deteriorates drastically when applied to other degradation settings. In this paper, we propose deep likelihood network (DL-Net), aiming at generalizing off-the-shelf image restoration networks to succeed over a spectrum of degradation levels. We slightly modify an off-the-shelf network by appending a simple recursive module, which is derived from a fidelity term, for disentangling the computation for multiple degradation levels. Extensive experimental results on image inpainting, interpolation, and super-resolution show the effectiveness of our DL-Net.

We propose a compact and effective framework to fuse multimodal features at multiple layers in a single network. The framework consists of two innovative fusion schemes. Firstly, unlike existing multimodal methods that necessitate individual encoders for different modalities, we verify that multimodal features can be learnt within a shared single network by merely maintaining modality-specific batch normalization layers in the encoder, which also enables implicit fusion via joint feature representation learning. Secondly, we propose a bidirectional multi-layer fusion scheme, where multimodal features can be exploited progressively. To take advantage of such scheme, we introduce two asymmetric fusion operations including channel shuffle and pixel shift, which learn different fused features with respect to different fusion directions. These two operations are parameter-free and strengthen the multimodal feature interactions across channels as well as enhance the spatial feature discrimination within channels. We conduct extensive experiments on semantic segmentation and image translation tasks, based on three publicly available datasets covering diverse modalities. Results indicate that our proposed framework is general, compact and is superior to state-of-the-art fusion frameworks.

Portrait relighting aims to render a face image under different lighting conditions. Existing methods do not explicitly consider some challenging lighting effects such as specular and shadow, and thus may fail in handling extreme lighting conditions. In this paper, we propose a novel framework that explicitly models multiple reflectance channels for single image portrait relighting, including the facial albedo, geometry as well as two lighting effects, i.e. , specular and shadow. These channels are finally composed to generate the relit results via deep neural networks. Current datasets do not support learning such multiple reflectance channel modeling. Therefore, we present a large-scale dataset with the ground-truths of the channels, enabling us to train the deep neural networks in a supervised manner. Furthermore, we develop a novel module named Lighting guided Feature Modulation (LFM). In contrast to existing methods which simply incorporate the given lighting in the bottleneck of a network, LFM fuses the lighting by layer-wise feature modulation to deliver more convincing results. Extensive experiments demonstrate that our proposed method achieves better results and is able to generate challenging lighting effects.

Universal style transfer tries to explicitly minimize the losses in feature space, thus it does not require training on any pre-defined styles. It usually uses different layers of VGG network as the encoders and trains several decoders to invert the features into images. Therefore, the effect of style transfer is achieved by feature transform. Although plenty of methods have been proposed, a theoretical analysis of feature transform is still missing. In this paper, we first propose a novel interpretation by treating it as the optimal transport problem. Then, we demonstrate the relations of our formulation with former works like Adaptive Instance Normalization (AdaIN) and Whitening and Coloring Transform (WCT). Finally, we derive a closed-form solution named Optimal Style Transfer (OST) under our formulation by additionally considering the content loss of Gatys. Comparatively, our solution can preserve better structure and achieve visually pleasing results. It is simple yet effective and we demonstrate its advantages both quantitatively and qualitatively. Besides, we hope our theoretical analysis can inspire future works in neural style transfer.

We propose a novel method for robust 6-DOF pose tracking of rigid objects from monocular images. In our method, 3D object tracking is achieved by directly aligning video frames to dynamic templates rendered from a textured 3D object model. Unlike previous methods, which usually utilize a small number of discrete templates to align with video frames, we employ an online textured model, rendering to create dynamic templates in continuous pose space according to the previously estimated object pose. In this way, a pose estimator could be easily converged to the optimal state. Besides, the rendered template also helps to detect the occlusion area by comparing it with the current frame, making our method highly robust to partial occlusions. The performance of our method is further improved by introducing a generic representation of dense images features, which we call extended dense feature fields (EDFF). Different kinds of pixel-level image features can be added to the EDFF and be optimized simultaneously in a unified Gauss-Newton optimization scheme. Attributing to dynamic templates from the textured model rendering and complementary features in EDFF, our method is able to deal with poor-textured and specular objects, as well as lighting variation and heavy occlusions. While our method is quite simple and straightforward, it achieves competitive or even superior results compared with the state of the art on challenging data sets.

We present RON, an efficient and effective framework for generic object detection. Our motivation is to smartly associate the best of the region-based (e.g., Faster R-CNN) and region-free (e.g., SSD) methodologies. Under fully convolutional architecture, RON mainly focuses on two fundamental problems: (a) multi-scale object localization and (b) negative sample mining. To address (a), we design the reverse connection, which enables the network to detect objects on multi-levels of CNNs. To deal with (b), we propose the objectness prior to significantly reduce the searching space of objects. We optimize the reverse connection, objectness prior and object detector jointly by a multi-task loss function, thus RON can directly predict final detection results from all locations of various feature maps. Extensive experiments on the challenging PASCAL VOC 2007, PASCAL VOC 2012 and MS COCO benchmarks demonstrate the competitive performance of RON. Specifically, with VGG-16 and low resolution 384×384 input size, the network gets 81.3% mAP on PASCAL VOC 2007, 80.7% mAP on PASCAL VOC 2012 datasets. Its superiority increases when datasets become larger and more difficult, as demonstrated by the results on the MS COCO dataset. With 1.5G GPU memory at test phase, the speed of the network is 15 FPS, 3 times faster than the Faster R-CNN counterpart. Code will be made publicly available.

In this paper, we propose an alternative method to estimate room layouts of cluttered indoor scenes. This method enjoys the benefits of two novel techniques. The first one is semantic transfer (ST), which is: (1) a formulation to integrate the relationship between scene clutter and room layout into convolutional neural networks, (2) an architecture that can be end-to-end trained, (3) a practical strategy to initialize weights for very deep networks under unbalanced training data distribution. ST allows us to extract highly robust features under various circumstances, and in order to address the computation redundance hidden in these features we develop a principled and efficient inference scheme named physics inspired optimization (PIO). PIOs basic idea is to formulate some phenomena observed in ST features into mechanics concepts. Evaluations on public datasets LSUN and Hedau show that the proposed method is more accurate than state-of-the-art methods.

Recently, the community of style transfer is trying to incorporate semantic information into traditional system. This practice achieves better perceptual results by transferring the style between semantically-corresponding regions. Yet, few efforts are invested to address the computation bottleneck of back-propagation. In this paper, we propose a new framework for fast semantic style transfer. Our method decomposes the semantic style transfer problem into feature reconstruction part and feature decoder part. The reconstruction part tactfully solves the optimization problem of content loss and style loss in feature space by particularly reconstructed feature. This significantly reduces the computation of propagating the loss through the whole network. The decoder part transforms the reconstructed feature into the stylized image. Through a careful bridging of the two modules, the proposed approach not only achieves competitive results as backward optimization methods but also is about two orders of magnitude faster.

State-of-the-art real-time face tracking systems still lack the ability to realistically portray subtle details of various aspects of the face, particularly the region surrounding the eyes. To improve this situation, we propose a technique to reconstruct the 3D shape and motion of eyelids in real time. By combining these results with the full facial expression and gaze direction, our system generates complete face tracking sequences with more detailed eye regions than existing solutions in real-time. To achieve this goal, we propose a generative eyelid model which decomposes eyelid variation into two low-dimensional linear spaces which efficiently represent the shape and motion of eyelids. Then, we modify a holistically-nested DNN model to jointly perform semantic eyelid edge detection and identification on images. Next, we correspond vertices of the eyelid model to 2D image edges, and employ polynomial curve fitting and a search scheme to handle incorrect and partial edge detections. Finally, we use the correspondences in a 3D-to-2D edge fitting scheme to reconstruct eyelid shape and pose. By integrating our fast fitting method into a face tracking system, the estimated eyelid results are seamlessly fused with the face and eyeball results in real time. Experiments show that our technique applies to different human races, eyelid shapes, and eyelid motions, and is robust to changes in head pose, expression and gaze direction.