Existing research on unconstrained in-the-wild head pose estimation suffers from the flaws of its datasets, which consist of either numerous samples by non-realistic synthesis or constrained collection, or small-scale natural images yet with plausible manual annotations. This makes fully-supervised solutions compromised due to the reliance on generous labels. To alleviate it, we propose the first semi-supervised unconstrained head pose estimation method SemiUHPE, which can leverage abundant easily available unlabeled head images. Technically, we choose semi-supervised rotation regression and adapt it to the error-sensitive and label-scarce problem of unconstrained head pose. Our method is based on the observation that the aspect-ratio invariant cropping of wild heads is superior to previous landmark-based affine alignment given that landmarks of unconstrained human heads are usually unavailable, especially for underexplored non-frontal heads. Instead of using a pre-fixed threshold to filter out pseudo labeled heads, we propose dynamic entropy based filtering to adaptively remove unlabeled outliers as training progresses by updating the threshold in multiple stages. We then revisit the design of weak-strong augmentations and improve it by devising two novel head-oriented strong augmentations, termed pose-irrelevant cut-occlusion and pose-altering rotation consistency respectively. Extensive experiments and ablation studies show that SemiUHPE outperforms its counterparts greatly on public benchmarks under both the front-range and full-range settings. Furthermore, our proposed method is also beneficial for solving other closely related problems, including generic object rotation regression and 3D head reconstruction, demonstrating good versatility and extensibility.

Bimanual robotic manipulation remains a fundamental challenge due to the inherent complexity of dual-arm coordination and high-dimensional action spaces. This paper presents the extended YOTO++ (You Only Teach Once), which is a unified one-shot learning framework for teaching bimanual skills directly from third-person human video demonstrations. Our method extracts structured 3D hand motions using binocular vision and distills them into compact, keyframe-based trajectories for dual-arm execution. We develop a scalable demonstration proliferation strategy that synthetically augments one-shot demonstrations into diverse training samples, enabling effective learning of a customized bimanual diffusion policy. Extensive evaluations across a broad spectrum of long-horizon bimanual tasks, including asynchronous, synchronous, contact-rich, and non-prehensile scenarios, demonstrate strong generalization to novel skills and objects. We further introduce a visual alignment mechanism at the initial manipulation stage for closed-loop control, enabling the system to timely adapt to perturbations during execution. We validate the framework on an unseen dual-arm robotic platform to show seamless cross-embodiment transfer without additional retraining. YOTO++ achieves impressive performance in accuracy, robustness, and scalability, advancing the practical deployment of general-purpose bimanual manipulation systems.

Learning semantic representations from point sets of 3D object shapes is often challenged by significant geometric variations, primarily due to differences in data acquisition methods. Typically, training data is generated using point simulators, while testing data is collected with distinct 3D sensors, leading to a simulation-to-reality (Sim2Real) domain gap that limits the generalization ability of point classifiers. Current unsupervised domain adaptation (UDA) techniques struggle with this gap, as they often lack robust, domain-insensitive descriptors capable of capturing global topological information, resulting in overfitting to the limited semantic patterns of the source domain. To address this issue, we introduce a novel Topology-Aware Modeling (TAM) framework for Sim2Real UDA on object point clouds. Our approach mitigates the domain gap by leveraging global spatial topology, characterized by low-level, high-frequency 3D structures, and by modeling the topological relations of local geometric features through a novel self-supervised learning task. Additionally, we propose an advanced self-training strategy that combines cross-domain contrastive learning with self-training, effectively reducing the impact of noisy pseudo-labels and enhancing the robustness of the adaptation process. Experimental results on three public Sim2Real benchmarks validate the effectiveness of our TAM framework, showing consistent improvements over state-of-the-art methods across all evaluated tasks. The source code of this work will be available athttps://github.com/zou-longkun/TAG.git.

In recent years, there has been a growing demand to stylize a given 3D scene to align with the artistic style of reference images for creative purposes. While 3D Gaussian Splatting (GS) has emerged as a promising and efficient method for realistic 3D scene modeling, there remains a challenge in adapting it to stylize 3D GS to match with multiple styles through automatic local style transfer or manual designation, while maintaining memory efficiency for stylization training. In this paper, we introduce a novel 3D GS stylization solution termed Multi-StyleGS to tackle these challenges. In particular, we employ a bipartite matching mechanism to automatically identify correspondences between the style images and the local regions of the rendered images. To facilitate local style transfer, we introduce a novel semantic style loss function that employs a segmentation network to apply distinct styles to various objects of the scene and propose a local-global feature matching to enhance the multi-view consistency. Furthermore, this technique can achieve memory-efficient training, more texture details and better color match. To better assign a robust semantic label to each Gaussian, we propose several techniques to regularize the segmentation network. As demonstrated by our comprehensive experiments, our approach outperforms existing ones in producing plausible stylization results and offering flexible editing.

Differentiable rendering has gained significant attention in the field of robotics, with differentiable robot rendering emerging as an effective paradigm for learning robotic actions from image-space supervision. However, the lack of physical world perception in this approach may lead to potential collisions during action optimization. In this work, we introduce a novel improvement on previous efforts by incorporating physical awareness of collisions through the learning of a neural robotic collision classifier. This enables the optimization of actions that avoid collisions with static, non-interactable environments as well as the robot itself. To facilitate effective gradient optimization with the classifier, we identify the underlying issue and propose leveraging Eikonal regularization to ensure consistent gradients for optimization. Our solution can be seamlessly integrated into existing differentiable robot rendering frameworks, utilizing gradients for optimization and providing a foundation for future applications of differentiable rendering in robotics with improved reliability of interactions with the physical world. Both qualitative and quantitative experiments demonstrate the necessity and effectiveness of our method compared to previous solutions.

Vision-language models (VLMs) have revolutionized machine learning by leveraging large pre-trained models to tackle various downstream tasks. Although label, training, and data efficiency have improved, many state-of-the-art VLMs still require task-specific hyperparameter tuning and fail to fully exploit test samples. To overcome these challenges, we propose a graph-based approach for label-efficient adaptation and inference. Our method dynamically constructs a graph over text prompts, few-shot examples, and test samples, using label propagation for inference without task-specific tuning. Unlike existing zero-shot label propagation techniques, our approach requires no additional unlabeled support set and effectively leverages the test sample manifold through dynamic graph expansion. We further introduce a context-aware feature re-weighting mechanism to improve task adaptation accuracy. Additionally, our method supports efficient graph expansion, enabling real-time inductive inference. Extensive evaluations on downstream tasks, such as fine-grained categorization and out-of-distribution generalization, demonstrate the effectiveness of our approach. The source code is available at https://github.com/Yushu-Li/ECALP.

Test-time adaptation (TTA) updates the model weights during the inference stage using testing data to enhance generalization. However, this practice exposes TTA to adversarial risks. Existing studies have shown that when TTA is updated with crafted adversarial test samples, also known as test-time poisoned data, the performance on benign samples can deteriorate. Nonetheless, the perceived adversarial risk may be overstated if the poisoned data is generated under overly strong assumptions. In this work, we first review realistic assumptions for test-time data poisoning, including white-box versus grey-box attacks, access to benign data, attack order, and more. We then propose an effective and realistic attack method that better produces poisoned samples without access to benign samples, and derive an effective in-distribution attack objective. We also design two TTA-aware attack objectives. Our benchmarks of existing attack methods reveal that the TTA methods are more robust than previously believed. In addition, we analyze effective defense strategies to help develop adversarially robust TTA methods. The source code is available at https://github.com/Gorilla-Lab-SCUT/RTTDP.

A critical prerequisite for achieving generalizable robot control is the availability of a large-scale robot training dataset. Due to the expense of collecting realistic robotic data, recent studies explored simulating and recording robot skills in virtual environments. While simulated data can be generated at higher speeds, lower costs, and larger scales, the applicability of such simulated data remains questionable due to the gap between simulated and realistic environments. To advance the Sim2Real generalization, in this study, we present DexScale, a data engine designed to perform automatic skills simulation and scaling for learning deployable robot manipulation policies. Specifically, DexScale ensures the usability of simulated skills by integrating diverse forms of realistic data into the simulated environment, preserving semantic alignment with the target tasks. For each simulated skill in the environment, DexScale facilitates effective Sim2Real data scaling by automating the process of domain randomization and adaptation. Tuned by the scaled dataset, the control policy achieves zero-shot Sim2Real generalization across diverse tasks, multiple robot embodiments, and widely studied policy model architectures, highlighting its importance in advancing Sim2Real embodied intelligence.

Achieving precise and generalizable grasping across diverse objects and environments is essential for intelligent and collaborative robotic systems. However, existing approaches often struggle with ambiguous affordance reasoning and limited adaptability to unseen objects, leading to suboptimal grasp execution. In this work, we propose GAT-Grasp, a gesture-driven grasping framework that directly utilizes human hand gestures to guide the generation of task-specific grasp poses with appropriate positioning and orientation. Specifically, we introduce a retrieval-based affordance transfer paradigm, leveraging the implicit correlation between hand gestures and object affordances to extract grasping knowledge from large-scale human-object interaction videos. By eliminating the reliance on pre-given object priors, GAT-Grasp enables zero-shot generalization to novel objects and cluttered environments. Real-World evaluations confirm its robustness across diverse and unseen scenarios, demonstrating reliable grasp execution in complex task settings.

Directly generating 3D meshes, the default representation for 3D shapes in the graphics industry, using auto-regressive (AR) models has become popular these days, thanks to their sharpness, compactness in the generated results, and ability to represent various types of surfaces. However, AR mesh generative models typically construct meshes face by face in lexicographic order, which does not effectively capture the underlying geometry in a manner consistent with human perception. Inspired by 2D models that progressively refine images, such as the prevailing next-scale prediction AR models, we propose generating meshes auto-regressively in a progressive coarse-to-fine manner. Specifically, we view mesh simplification algorithms, which gradually merge mesh faces to build simpler meshes, as a natural fine-to-coarse process. Therefore, we generalize meshes to simplicial complexes and develop a transformer-based AR model to approximate the reverse process of simplification in the order of level of detail, constructing meshes initially from a single point and gradually adding geometric details through local remeshing, where the topology is not predefined and is alterable. Our experiments show that this novel progressive mesh generation approach not only provides intuitive control over generation quality and time consumption by early stopping the auto-regressive process but also enables applications such as mesh refinement and editing.

Generating high-quality and photorealistic 3D assets remains a longstanding challenge in 3D vision and computer graphics. Although state-of-the-art generative models, such as diffusion models, have made significant progress in 3D generation, they often fall short of human-designed content due to limited ability to follow instructions, align with human preferences, or produce realistic textures, geometries, and physical attributes. In this paper, we introduce Nabla-R2D3, a highly effective and sample-efficient reinforcement learning alignment framework for 3D-native diffusion models using 2D rewards. Built upon the recently proposed Nabla-GFlowNet method for reward finetuning, our Nabla-R2D3 enables effective adaptation of 3D diffusion models through pure 2D reward feedback. Extensive experiments show that, unlike naive finetuning baselines which either fail to converge or suffer from overfitting, Nabla-R2D3 consistently achieves higher rewards and reduced prior forgetting within few finetuning steps.

Humans can easily deduce the relative pose of a previously unseen object, without labeling or training, given only a single query-reference image pair. This is arguably achieved by incorporating i) 3D/2.5D shape perception from a single image, ii) render-and-compare simulation, and iii) rich semantic cue awareness to furnish (coarse) reference-query correspondence. Motivated by this, we propose a novel 3D generalizable relative pose estimation method by elaborating 3D/2.5D shape perception with a 2.5D shape from an RGB-D reference, fulfilling the render-and-compare paradigm with an off-the-shelf differentiable renderer, and leveraging the semantic cues from a pretrained model like DINOv2. Specifically, our differentiable renderer takes the 2.5D rotatable mesh textured by the RGB and the semantic maps (obtained by DINOv2 from the RGB input), then renders new RGB and semantic maps (with back-surface culling) under a novel rotated view. The refinement loss comes from comparing the rendered RGB and semantic maps with the query ones, back-propagating the gradients through the differentiable renderer to refine the 3D relative pose. As a result, our method can be readily applied to unseen objects, given only a single RGB-D reference, without labeling or training. Extensive experiments on LineMOD, LM-O, and YCB-V show that our training-free method significantly outperforms the state-of-the-art supervised methods, especially under the rigorous Acc@5/10/15$^\circ$∘ metrics and the challenging cross-dataset settings.

Recovering the intrinsic physical attributes of a scene from images, generally termed as the inverse rendering problem, has been a central and challenging task in computer vision and computer graphics. In this paper, we present GUS-IR, a novel framework designed to address the inverse rendering problem for complicated scenes featuring rough and glossy surfaces. This paper starts by analyzing and comparing two prominent shading techniques popularly used for inverse rendering, forward shading and deferred shading, effectiveness in handling complex materials. More importantly, we propose a unified shading solution that combines the advantages of both techniques for better decomposition. In addition, we analyze the normal modeling in 3D Gaussian Splatting (3DGS) and utilize the shortest axis as normal for each particle in GUS-IR, along with a depth-related regularization, resulting in improved geometric representation and better shape reconstruction. Furthermore, we enhance the probe-based baking scheme proposed by GS-IR to achieve more accurate ambient occlusion modeling to better handle indirect illumination. Extensive experiments have demonstrated the superior performance of GUS-IR in achieving precise intrinsic decomposition and geometric representation, supporting many downstream tasks (such as relighting, retouching) in computer vision, graphics, and extended reality.

Modeling indirect illumination to handle global illumination and decompose materials from multi-view images is challenging, especially in complex scenes with self-occlusion. While recent implicit neural representations show promise in inverse rendering, they struggle with efficient and effective modeling of indirect illumination. Besides, real-time global illumination techniques (e.g. Indirect Lighting Cache) have been successful in gaming. Inspired by this, we present a novel three-stageInverseRenderer withProbes (IR-Pro), which efficiently caches occlusion to handle indirect illumination. Experiments demonstrate the superiority of IR-Pro over existing methods in the inverse rendering of complex scenes. Furthermore, we successfully integrate the results into digital content creation software and showcase their effectiveness in applications, like relighting, simulation, and editing.

Masked autoencoder (MAE) has demonstrated its effectiveness in self-supervised point cloud learning. Considering that masking is a kind of corruption, in this work we explore a more general denoising autoencoder for point cloud learning (Point-DAE) by investigating more types of corruptions beyond masking. Specifically, we degrade the point cloud with certain corruptions as input, and learn an encoder-decoder model to reconstruct the original point cloud from its corrupted version. Three corruption families (i.e., density/masking, noise, and affine transformation) and a total of 14 corruption types are investigated with traditional non-Transformer encoders. Besides the popular masking corruption, we identify another effective corruption family, i.e., affine transformation. The affine transformation disturbs all points globally, which is complementary to the masking corruption where some local regions are dropped. We also validate the effectiveness of affine transformation corruption with the Transformer backbones, where we decompose the reconstruction of the complete point cloud into the reconstructions of detailed local patches and rough global shape, alleviating the position leakage problem in the reconstruction. Extensive experiments on tasks of object classification, few-shot learning, robustness testing, part segmentation, and 3-D object detection validate the effectiveness of the proposed method. The codes are available at https://github.com/YBZh/Point-DAE.

Test-Time Adaptation aims to adapt source domain model to testing data at inference stage with success demonstrated in adapting to unseen corruptions. However, these attempts may fail under more challenging real-world scenarios. Existing works mainly consider real-world test-time adaptation under non-i.i.d. data stream and continual domain shift. In this work, we first complement the existing real-world TTA protocol with a globally class imbalanced testing set. We demonstrate that combining all settings together poses new challenges to existing methods. We argue the failure of state-of-the-art methods is first caused by indiscriminately adapting normalization layers to imbalanced testing data. To remedy this shortcoming, we propose a balanced batchnorm layer to swap out the regular batchnorm at inference stage. The new batchnorm layer is capable of adapting without biasing towards majority classes. We are further inspired by the success of self-training (ST) in learning from unlabeled data and adapt ST for test-time adaptation. However, ST alone is prone to over adaption which is responsible for the poor performance under continual domain shift. Hence, we propose to improve self-training under continual domain shift by regularizing model updates with an anchored loss. The final TTA model, termed as TRIBE, is built upon a tri-net architecture with balanced batchnorm layers. We evaluate TRIBE on four datasets representing real-world TTA settings. TRIBE consistently achieves the state-of-the-art performance across multiple evaluation protocols. The code is available at https://github.com/Gorilla-Lab-SCUT/TRIBE.

We propose GS-IR, a novel inverse rendering approach based on 3D Gaussian Splatting (3DGS) that leverages forward mapping volume rendering to achieve photorealistic novel view synthesis and relighting results. Unlike previous works that use implicit neural representations and volume rendering (e.g. NeRF), which suffer from low expressive power and high computational complexity, we extend 3DGS, a top-performance representation for novel view synthesis, to estimate scene geometry, surface material, and environment illumination from multi-view images captured under unknown lighting conditions. There are two main problems when introducing 3DGS to inverse rendering: 1) 3DGS does not support producing plausible normal natively; 2) forward mapping (e.g. rasterization and splatting) cannot trace the occlusion like backward mapping (e.g. ray tracing). To address these challenges, our GS-IR proposes an efficient optimization scheme incorporating a depth-derivation-based regularization for normal estimation and a baking-based occlusion to model indirect lighting. The flexible and expressive 3DGS representation allows us to achieve fast and compact geometry reconstruction, photore-alistic novel view synthesis, and effective physically-based rendering. We demonstrate the superiority of our method over baseline methods through qualitative and quantitative evaluations of various challenging scenes. The source code is available at https://github.com/lzhnb/GS-IR.

Zero-shot 6D object pose estimation involves the detection of novel objects with their 6D poses in cluttered scenes, presenting significant challenges for model generalizability. Fortunately, the recent Segment Anything Model (SAM) has showcased remarkable zero-shot transfer performance, which provides a promising solution to tackle this task. Motivated by this, we introduce SAM-6D, a novel framework designed to realize the task through two steps, including instance segmentation and pose estimation. Given the target objects, SAM-6D employs two dedicated sub-networks, namely Instance Segmentation Model (ISM) and Pose Estimation Model (PEM), to perform these steps on cluttered RGB-D images. ISM takes SAM as an advanced starting point to generate all possible object proposals and selectively preserves valid ones through meticulously crafted object matching scores in terms of semantics, appearance and geometry. By treating pose estimation as a partial-to-partial point matching problem, PEM performs a two-stage point matching process featuring a novel design of background tokens to construct dense 3D-3D correspondence, ultimately yielding the pose estimates. Without bells and whistles, SAM-6D outperforms the existing methods on the seven core datasets of the BOP Benchmark for both instance segmentation and pose estimation of novel objects.

Deep Text-to-Image Synthesis (TIS) models such as Stable Diffusion have recently gained significant popularity for creative text-to-image generation. However, for domain-specific scenarios, tuning-free Text-guided Image Editing (TIE) is of greater importance for application developers. This approach modifies objects or object properties in images by manipulating feature components in attention layers during the generation process. Nevertheless, little is known about the semantic meanings that these attention layers have learned and which parts of the attention maps contribute to the success of image editing. In this paper, we conduct an in-depth probing analysis and demonstrate that cross-attention maps in Stable Diffusion often contain object attribution information, which can result in editing failures. In contrast, self-attention maps play a crucial role in preserving the geometric and shape details of the source image during the transformation to the target image. Our analysis offers valuable insights into understanding cross and self-attention mechanisms in diffusion models. Furthermore, based on our findings, we propose a simplified, yet more stable and efficient, tuning-free procedure that modifies only the self-attention maps of specified attention layers during the denoising process. Experimental results show that our simplified method consistently surpasses the performance of popular approaches on multiple datasets.11Source code and datasets are available at https://github.com/alibaba/EasyNLP/tree/master/diffusion/FreePromptEditing.

The success of large language models has inspired the computer vision community to explore image segmentation foundation model that is able to zero/few-shot generalize through prompt engineering. Segment-Anything (SAM), among others, is the state-of-the-art image segmentation foundation model demonstrating strong zero/few-shot generalization. Despite the success, recent studies reveal the weakness of SAM under strong distribution shift. In particular, SAM performs awkwardly on corrupted natural images, camouflaged images, medical images, etc. Motivated by the observations, we aim to develop a self-training based strategy to adapt SAM to target distribution. Given the unique challenges of large source dataset, high computation cost and incorrect pseudo label, we propose a weakly supervised self-training architecture with anchor regularization and low-rank finetuning to improve the robustness and computation efficiency of adaptation. We validate the effectiveness on 5 types of downstream segmentation tasks including natural clean/corrupted images, medical images, camouflaged images and robotic images. Our proposed method is task-agnostic in nature and outperforms pre-trained SAM and state-of-the-art domain adaptation methods on almost all downstream tasks with the same testing prompt inputs.

Building on recent breakthroughs in diffusion-based text-to-image synthesis (TIS), training-free text-guided image editing (TIE) has emerged as an indispensable aspect of modern image editing practices. This technique involves the modification of features within attention layers to alter objects or their attributes within images during the generation process. Despite its utility, current image editing algorithms face challenges, particularly when editing multiple objects in an image. In this paper, we introduce VICTORIA, a novel approach that augments TIE by incorporating linguistic knowledge into the manipulation of attention maps during image generation. VICTORIA capitalizes on mechanisms within self-attention layers to ensure spatial consistency between source and target images. Further, we design a novel loss function that refines cross-attention maps, ensuring their alignment with linguistic constraints, thereby enhancing the editing precision of multiple target objects. We also present a linguistic mask blending technique that aids in the retention of information in regions not subjected to modification. Experimental results across seven diverse datasets show that VICTORIA achieves significant improvements over state-of-the-art methods. Our work underscores the critical role and effectiveness of linguistic analysis in elevating the performance of TIE, with a specific emphasis on multi-object scenarios. The code is available at https://github.com/alibaba/EasyNLP/tree/master/diffusion/VICTORIA.

Reconstruction of a continuous surface of two-dimensional manifold from its raw, discrete point cloud observation is a long-standing problem in computer vision and graphics research. The problem is technically ill-posed, and becomes more difficult considering that various sensing imperfections would appear in the point clouds obtained by practical depth scanning. In literature, a rich set of methods has been proposed, and reviews of existing methods are also provided. However, existing reviews are short of thorough investigations on a common benchmark. The present paper aims to review and benchmark existing methods in the new era of deep learning surface reconstruction. To this end, we contribute a large-scale benchmarking dataset consisting of both synthetic and real-scanned data; the benchmark includes object- and scene-level surfaces and takes into account various sensing imperfections that are commonly encountered in practical depth scanning. We conduct thorough empirical studies by comparing existing methods on the constructed benchmark, and pay special attention on robustness of existing methods against various scanning imperfections; we also study how different methods generalize in terms of reconstructing complex surface shapes. Our studies help identity the best conditions under which different methods work, and suggest some empirical findings. For example, while deep learning methods are increasingly popular in the research community, our systematic studies suggest that, surprisingly, a few classical methods perform even better in terms of both robustness and generalization; our studies also suggest that the practical challenges of misalignment of point sets from multi-view scanning, missing of surface points, and point outliers remain unsolved by all the existing surface reconstruction methods. We expect that the benchmark and our studies would be valuable both for practitioners and as a guidance for new innovations in future research.

Deploying models on target domain data subject to distribution shift requires adaptation. Test-time training (TTT) emerges as a solution to this adaptation under a realistic scenario where access to full source domain data is not available, and instant inference on the target domain is required. Despite many efforts into TTT, there is a confusion over the experimental settings, thus leading to unfair comparisons. In this work, we first revisit TTT assumptions and categorize TTT protocols by two key factors, i.e., whether testing data is sequentially streamed and whether source model is allowed to be trained with modified loss function. Among the multiple protocols, we adopt a realistic sequential test-time training (sTTT) protocol, under which we develop a test-time anchored clustering (TTAC) approach to enable stronger test-time feature learning. TTAC discovers clusters in both source and target domains and matches the target clusters to the source ones to improve adaptation. When source domain information is strictly absent (i.e., source-free) we further develop an efficient method to infer source domain distributions for anchored clustering. Finally, self-training (ST) has demonstrated great success in learning from unlabeled data and we empirically figure out that applying ST alone to TTT is prone to confirmation bias. Therefore, a more effective TTT approach is introduced by regularizing self-training with anchored clustering, and the improved model is referred to as TTAC++. We demonstrate that, under all TTT protocols, TTAC++ consistently outperforms the state-of-the-art methods on five TTT datasets, including corrupted target domain, selected hard samples, synthetic-to-real adaptation and adversarially attacked target domain. We hope this work will provide a fair benchmarking of TTT methods, and future research should be compared within respective protocols.

We study multi-sensor fusion for 3D semantic segmentation that is important to scene understanding for many applications, such as autonomous driving and robotics. For example, for autonomous cars equipped with RGB cameras and LiDAR, it is crucial to fuse complementary information from different sensors for robust and accurate segmentation. Existing fusion-based methods, however, may not achieve promising performance due to the vast difference between the two modalities. In this work, we investigate a collaborative fusion scheme called perception-aware multi-sensor fusion (PMF) to effectively exploit perceptual information from two modalities, namely, appearance information from RGB images and spatio-depth information from point clouds. To this end, we first project point clouds to the camera coordinate using perspective projection. In this way, we can process both inputs from LiDAR and cameras in 2D space while preventing the information loss of RGB images. Then, we propose a two-stream network that consists of a LiDAR stream and a camera stream to extract features from the two modalities, separately. The extracted features are fused by effective residual-based fusion modules. Moreover, we introduce additional perception-aware losses to measure the perceptual difference between the two modalities. Last, we propose an improved version of PMF,i.e., EPMF, which is more efficient and effective by optimizing data pre-processing and network architecture under perspective projection. Specifically, we propose cross-modal alignment and cropping to obtain tight inputs and reduce unnecessary computational costs. We then explore more efficient contextual modules under perspective projection and fuse the LiDAR features into the camera stream to boost the performance of the two-stream network. Extensive experiments on benchmark data sets show the superiority of our method. For example, on nuScenes test set, our EPMF outperforms the state-of-the-art method,i.e., RangeFormer, by0.9%in mIoU. Compared to PMF, EPMF also achieves2.06× acceleration with2.0%improvement in mIoU. Our source code is available athttps://github.com/ICEORY/PMF.

Semantic pattern of an object point cloud is determined by its topological configuration of local geometries. Learning discriminative representations can be challenging due to large shape variations of point sets in local regions and incomplete surface in a global perspective, which can be made even more severe in the context of unsupervised domain adaptation (UDA). In specific, traditional 3D networks mainly focus on local geometric details and ignore the topological structure between local geometries, which greatly limits their cross-domain generalization. Recently, the transformer-based models have achieved impressive performance gain in a range of image-based tasks, benefiting from its strong generalization capability and scalability stemming from capturing long range correlation across local patches. Inspired by such successes of visual transformers, we propose a novel Relational Priors Distillation (RPD) method to extract relational priors from the well-trained transformers on massive images, which can significantly empower cross-domain representations with consistent topological priors of objects. To this end, we establish a parameter-frozen pre-trained transformer module shared between 2D teacher and 3D student models, complemented by an online knowledge distillation strategy for semantically regularizing the 3D student model. Furthermore, we introduce a novel self-supervised task centered on reconstructing masked point cloud patches using corresponding masked multi-view image features, thereby empowering the model with incorporating 3D geometric information. Experiments on the PointDA-10 and the Sim-to-Real datasets verify that the proposed method consistently achieves the state-of-the-art performance of UDA for point cloud classification. The source code of this work is available athttps://github.com/zou-longkun/RPD.git.

3D object detection has recently received much attention due to its great potential in autonomous vehicle (AV). The success of deep learning based object detectors relies on the availability of large-scale annotated datasets, which is time-consuming and expensive to compile, especially for 3D bounding box annotation. In this work, we investigate diversity-based active learning (AL) as a potential solution to alleviate the annotation burden. Given limited annotation budget, only the most informative frames and objects are automatically selected for human to annotate. Technically, we take the advantage of the multimodal information provided in an AV dataset, and propose a novel acquisition function that enforces spatial and temporal diversity in the selected samples. We benchmark the proposed method against other AL strategies under realistic annotation cost measurements, where the realistic costs for annotating a frame and a 3D bounding box are both taken into consideration. We demonstrate the effectiveness of the proposed method on the nuScenes dataset and show that it outperforms existing AL strategies significantly.

Domain generalization (DG) for person re-identification (ReID) is a challenging problem, as access to target domain data is not permitted during the training process. Most existing DG ReID methods update the feature extractor and classifier parameters based on the same features. This common practice causes the model to overfit to existing feature styles in the source domain, resulting in sub-optimal generalization ability on target domains. To solve this problem, we propose a novel style interleaved learning (IL) framework. Unlike conventional learning strategies, IL incorporates two forward propagations and one backward propagation for each iteration. We employ the features of interleaved styles to update the feature extractor and classifiers using different forward propagations, which helps to prevent the model from overfitting to certain domain styles. To generate interleaved feature styles, we further propose a new feature stylization approach. It produces a wide range of meaningful styles that are both different and independent from the original styles in the source domain, which caters to the IL methodology. Extensive experimental results show that our model not only consistently outperforms state-of-the-art methods on large-scale benchmarks for DG ReID, but also has clear advantages in computational efficiency. The code is available athttps://github.com/WentaoTan/Interleaved-Learning.

We address the challenge of recovering an underlying scene geometry and colors from a sparse set of RGBD view observations. In this work, we present a new solution termed RGBD2that sequentially generates novel RGBD views along a camera trajectory, and the scene geometry is simply the fusion result of these views. More specifically, we maintain an intermediate surface mesh used for rendering new RGBD views, which subsequently becomes complete by an inpainting network; each rendered RGBD view is later back-projected as a partial surface and is supplemented into the intermediate mesh. The use of intermediate mesh and camera projection helps solve the tough problem of multi-view inconsistency. We practically implement the RGBD inpainting network as a versatile RGBD diffusion model, which is previously used for 2D generative modeling; we make a modification to its reverse diffusion process to enable our use. We evaluate our approach on the task of 3D scene synthesis from sparse RGBD inputs; extensive experiments on the ScanNet dataset demonstrate the superiority of our approach over existing ones. Project page: https://jblei.site/proj/rgbd-diffusion.

Recovery of an underlying scene geometry from multi-view images stands as a long-time challenge in computer vision research. The recent promise leverages neural implicit surface learning and differentiable volume rendering, and achieves both the recovery of scene geometry and synthesis of novel views, where deep priors of neural models are used as an inductive smoothness bias. While promising for object-level surfaces, these methods suffer when coping with complex scene surfaces. In the meanwhile, traditional multi-view stereo can recover the geometry of scenes with rich textures, by globally optimizing the local, pixel-wise correspondences across multiple views. We are thus motivated to make use of the complementary benefits from the two strategies, and propose a method termed Helix-shaped neural implicit Surface learning or HelixSurf; HelixSurf uses the intermediate prediction from one strategy as the guidance to regularize the learning of the other one, and conducts such intertwined regularization iteratively during the learning process. We also propose an efficient scheme for differentiable volume rendering in HelixSurf Experiments on surface reconstruction of indoor scenes show that our method compares favorably with existing methods and is orders of magnitude faster, even when some of existing methods are assisted with auxiliary training data. The source code is available at hups://guhub.com/Gorilla-Lab-SCUT/HelixSurf.

Deep learning in computer vision has achieved great success with the price of large-scale labeled training data. However, exhaustive data annotation is impracticable for each task of all domains of interest, due to high labor costs and unguaranteed labeling accuracy. Besides, the uncontrollable data collection process produces non-IID training and test data, where undesired duplication may exist. All these nuisances may hinder the verification of typical theories and exposure to new findings. To circumvent them, an alternative is to generate synthetic data via 3D rendering with domain randomization. We in this work push forward along this line by doing profound and extensive research on bare supervised learning and downstream domain adaptation. Specifically, under the well-controlled, IID data setting enabled by 3D rendering, we systematically verify the typical, important learning insights, e.g., shortcut learning, and discover the new laws of various data regimes and network architectures in generalization. We further investigate the effect of image formation factors on generalization, e.g., object scale, material texture, illumination, camera view-point, and background in a 3D scene. Moreover, we use the simulation-to-reality adaptation as a downstream task for comparing the transferability between synthetic and real data when used for pre-training, which demonstrates that synthetic data pre-training is also promising to improve real test results. Lastly, to promote future research, we develop a new large-scale synthetic-to-real benchmark for image classification, termed S2RDA, which provides more significant challenges for transfer from simulation to reality.

Automatic 3D content creation has achieved rapid progress recently due to the availability of pre-trained, large language models and image diffusion models, forming the emerging topic of text-to-3D content creation. Existing text-to-3D methods commonly use implicit scene representations, which couple the geometry and appearance via volume rendering and are suboptimal in terms of recovering finer geometries and achieving photorealistic rendering; consequently, they are less effective for generating high-quality 3D assets. In this work, we propose a new method of Fantasia3D for high-quality text-to-3D content creation. Key to Fantasia3D is the disentangled modeling and learning of geometry and appearance. For geometry learning, we rely on a hybrid scene representation, and propose to encode surface normal extracted from the representation as the input of the image diffusion model. For appearance modeling, we introduce the spatially varying bidirectional reflectance distribution function (BRDF) into the text-to-3D task, and learn the surface material for photorealistic rendering of the generated surface. Our disentangled framework is more compatible with popular graphics engines, supporting relighting, editing, and physical simulation of the generated 3D assets. We conduct thorough experiments that show the advantages of our method over existing ones under different text-to-3D task settings. Project page and source codes: https://fantasia3d.github.io/.

Generalizing deep learning models to unknown target domain distribution with low latency has motivated research into test-time training/adaptation (TTT/TTA). Existing approaches often focus on improving test-time training performance under well-curated target domain data. As figured out in this work, many state-of-the-art methods fail to maintain the performance when the target domain is contaminated with strong out-of-distribution (OOD) data, a.k.a. open-world test-time training (OWTTT). The failure is mainly due to the inability to distinguish strong OOD samples from regular weak OOD samples. To improve the robustness of OWTTT we first develop an adaptive strong OOD pruning which improves the efficacy of the self-training TTT method. We further propose a way to dynamically expand the prototypes to represent strong OOD samples for an improved weak/strong OOD data separation. Finally, we regularize self-training with distribution alignment and the combination yields the state-of-the-art performance on 5 OWTTT benchmarks. The code is available at https://github.com/Yushu-Li/OWTTT.

Rotation estimation of high precision from an RGB-D object observation is a huge challenge in 6D object pose estimation, due to the difficulty of learning in the non-linear space of SO (3). In this paper, we propose a novel rotation estimation network, termed as VI-Net, to make the task easier by decoupling the rotation as the combination of a viewpoint rotation and an in-plane rotation. More specifically, VI-Net bases the feature learning on the sphere with two individual branches for the estimates of two factorized rotations, where a V-Branch is employed to learn the viewpoint rotation via binary classification on the spherical signals, while another I-Branch is used to estimate the in-plane rotation by transforming the signals to view from the zenith direction. To process the spherical signals, a Spherical Feature Pyramid Network is constructed based on a novel design of SPAtial Spherical Convolution (SPA-SConv), which settles the boundary problem of spherical signals via feature padding and realizes viewpoint-equivariant feature extraction by symmetric convolutional operations. We apply the proposed VI-Net to the challenging task of category-level 6D object pose estimation for predicting the poses of unknown objects without available CAD models; experiments on the benchmarking datasets confirm the efficacy of our method, which outperforms the existing ones with a large margin in the regime of high precision.

Domain gap between synthetic and real data in visual regression (e.g., 6D pose estimation) is bridged in this paper via global feature alignment and local refinement on the coarse classification of discretized anchor classes in target space, which imposes a piece-wise target manifold regularization into domain-invariant representation learning. Specifically, our method incorporates an explicit self-supervised manifold regularization, revealing consistent cumulative target dependency across domains, to a self-training scheme (e.g., the popular Self-Paced Self-Training) to encourage more discriminative transferable representations of regression tasks. Moreover, learning unified implicit neural functions to estimate relative direction and distance of targets to their nearest class bins aims to refine target classification predictions, which can gain robust performance against inconsistent feature scaling sensitive to UDA regressors. Experiment results on three public benchmarks of the challenging 6D pose estimation task can verify the effectiveness of our method, consistently achieving superior performance to the state-of-the-art for UDA on 6D pose estimation. Codes and pre-trained models are available https://github.com/Gorilla-Lab-SCUT/MAST.

Addressing the annotation challenge in 3D Point Cloud segmentation has inspired research into weakly supervised learning. Existing approaches mainly focus on exploiting manifold and pseudo-labeling to make use of large unlabeled data points. A fundamental challenge here lies in the large intra-class variations of local geometric structure, resulting in subclasses within a semantic class. In this work, we leverage this intuition and opt for maintaining an individual classifier for each subclass. Technically, we design a multi-prototype classifier, each prototype serves as the classifier weights for one subclass. To enable effective updating of multi-prototype classifier weights, we propose two constraints respectively for updating the prototypes w.r.t. all point features and for encouraging the learning of diverse prototypes. Experiments on weakly supervised 3D point cloud segmentation tasks validate the efficacy of proposed method in particular at low-label regime. Our hypothesis is also verified given the consistent discovery of semantic subclasses at no cost of additional annotations.

Unsupervised Domain Adaptive (UDA) object re-identification (Re-ID) aims at adapting a model trained on a labeled source domain to an unlabeled target domain. State-of-the-art object Re-ID approaches adopt clustering algorithms to generate pseudo-labels for the unlabeled target domain. However, the inevitable label noise caused by the clustering procedure significantly degrades the discriminative power of Re-ID model. To address this problem, we propose an uncertainty-aware clustering framework (UCF) for UDA tasks. First, a novel hierarchical clustering scheme is proposed to promote clustering quality. Second, an uncertainty-aware collaborative instance selection method is introduced to select images with reliable labels for model training. Combining both techniques effectively reduces the impact of noisy labels. In addition, we introduce a strong baseline that features a compact contrastive loss. Our UCF method consistently achieves state-of-the-art performance in multiple UDA tasks for object Re-ID, and significantly reduces the performance gap between unsupervised and supervised Re-ID. In particular, the performance of our unsupervised UCF method in the MSMT17$\to$Market1501 task is better than that of the fully supervised setting on Market1501. The code of UCF is available athttps://github.com/Wang-pengfei/UCF.

Detecting objects from LiDAR point clouds is of tremendous significance in autonomous driving. In spite of good progress, accurate and reliable 3D detection is yet to be achieved due to the sparsity and irregularity of LiDAR point clouds. Among existing strategies, multi-view methods have shown great promise by leveraging the more comprehensive information from both bird's eye view (BEV) and range view (RV). These multi-view methods either refine the proposals predicted from single view via fused features, or fuse the features without considering the global spatial con-text; their performance is limited consequently. In this paper, we propose to adaptively fuse multi-view features in a global spatial context via Dual Cross- VIew SpaTial Attention (VISTA). The proposed VISTA is a novel plug-and-play fusion module, wherein the multi-layer perceptron widely adopted in standard attention modules is replaced with a convolutional one. Thanks to the learned attention mechanism, VISTA can produce fused features of high quality for prediction of proposals. We decouple the classification and regression tasks in VISTA, and an additional constraint of attention variance is applied that enables the attention module to focus on specific targets instead of generic points. We conduct thorough experiments on the benchmarks of nuScenes and Waymo; results confirm the efficacy of our designs. At the time of submission, our method achieves 63.0% in overall mAP and 69.8% in NDS on the nuScenes benchmark, outperforming all published methods by up to 24% in safety-crucial categories such as cyclist.

Semi-supervised learning (SSL) has been studied for a long time to solve vision tasks in data-efficient application scenarios. SSL aims to learn a good classification model using a few labeled data together with large-scale unlabeled data. Recent advances achieve the goal by combining multiple SSL techniques, e.g., self-training and consistency regularization. From unlabeled samples, they usually adopt a confidence filter (CF) to select reliable ones with high prediction confidence. In this work, we study whether the moderately confident samples are useless and how to select the useful ones to improve model optimization. To answer these problems, we propose a novel Taylor expansion inspired filtration (TEIF) framework, which admits the samples of moderate confidence with similar feature or gradient to the respective one averaged over the labeled and highly confident unlabeled data. It can produce a stable and new information induced network update, leading to better generalization. Two novel filters are derived from this framework and can be naturally explained in two perspectives. One is gradient synchronization filter (GSF), which strengthens the optimization dynamic of fully-supervised learning; it selects the samples whose gradients are similar to classwise majority gradients. The other is prototype proximity filter (PPF), which involves more prototypical samples in training to learn better semantic representations; it selects the samples near classwise prototypes. They can be integrated into SSL methods with CF. We use the state-of-the-art Fix-Match as the baseline. Experiments on popular SSL benchmarks show that we achieve the new state of the art.

Arbitrary style transfer (AST) and domain generalization (DG) are important yet challenging visual learning tasks, which can be cast as a feature distribution matching problem. With the assumption of Gaussian feature distribution, conventional feature distribution matching methods usually match the mean and standard deviation of features. However, the feature distributions of real-world data are usually much more complicated than Gaussian, which cannot be accurately matched by using only the first-order and second-order statistics, while it is computationally prohibitive to use high-order statistics for distribution matching. In this work, we, for the first time to our best knowledge, propose to perform Exact Feature Distribution Matching (EFDM) by exactly matching the empirical Cumulative Distribution Functions (eCDFs) of image features, which could be implemented by applying the Exact Histogram Matching (EHM) in the image feature space. Particularly, a fast EHM algorithm, named Sort-Matching, is employed to perform EFDM in a plug-and-play manner with minimal cost. The effectiveness of our proposed EFDM method is verified on a variety of AST and DG tasks, demonstrating new state-of-the-art results. Codes are available at https://github.com/YBZh/EFDM.

Semantic understanding of 3D point cloud relies on learning models with massively annotated data, which, in many cases, are expensive or difficult to collect. This has led to an emerging research interest in semi-supervised learning (SSL) for 3D point cloud. It is commonly assumed in SSL that the unlabeled data are drawn from the same distribution as that of the labeled ones; This assumption, however, rarely holds true in realistic environments. Blindly using out-of-distribution (OOD) unlabeled data could harm SSL performance. In this work, we propose to selectively utilize unlabeled data through sample weighting, so that only conducive unlabeled data would be prioritized. To estimate the weights, we adopt a bi-level optimization framework which iteratively optimizes a meta-objective on a held-out validation set and a task-objective on a training set. Faced with the instability of efficient bi-level optimizers, we further propose three regularization techniques to enhance the training stability. Extensive experiments on 3D point cloud classification and segmentation tasks verify the effectiveness of our proposed method. We also demonstrate the feasibility of a more efficient training strategy. Our code is released on Github1.

The challenges of learning a robust 6D pose function lie in 1) severe occlusion and 2) systematic noises in depth images. Inspired by the success of point-pair features, the goal of this paper is to recover the 6D pose of an object instance segmented from RGB-D images by locally matching pairs of oriented points between the model and camera space. To this end, we propose a novel Bi-directional Correspondence Mapping Network (BiCo-Net) to first generate point clouds guided by a typical pose regression, which can thus incorporate pose-sensitive information to optimize generation of local coordinates and their normal vectors. As pose predictions via geometric computation only rely on one single pair of local oriented points, our BiCo-Net can achieve robustness against sparse and occluded point clouds. An ensemble of redundant pose predictions from locally matching and direct pose regression further refines final pose output against noisy observations. Experimental results on three popularly benchmarking datasets can verify that our method can achieve state-of-the-art performance, especially for the more challenging severe occluded scenes. Source codes are available at https://github.com/Gorilla-Lab-SCUT/BiCo-Net.

Creation of 3D content by stylization is a promising yet challenging problem in computer vision and graphics research. In this work, we focus on stylizing photorealistic appearance renderings of a given surface mesh of arbitrary topology. Motivated by the recent surge of cross-modal supervision of the Contrastive Language-Image Pre-training (CLIP) model, we propose TANGO, which transfers the appearance style of a given 3D shape according to a text prompt in a photorealistic manner. Technically, we propose to disentangle the appearance style as the spatially varying bidirectional reflectance distribution function, the local geometric variation, and the lighting condition, which are jointly optimized, via supervision of the CLIP loss, by a spherical Gaussians based differentiable renderer. As such, TANGO enables photorealistic 3D style transfer by automatically predicting reflectance effects even for bare, low-quality meshes, without training on a task-specific dataset. Extensive experiments show that TANGO outperforms existing methods of text-driven 3D style transfer in terms of photorealistic quality, consistency of 3D geometry, and robustness when stylizing low-quality meshes. Our codes and results are available at our project webpage https://cyw-3d.github.io/tango/.

Deploying models on target domain data subject to distribution shift requires adaptation. Test-time training (TTT) emerges as a solution to this adaptation under a realistic scenario where access to full source domain data is not available and instant inference on target domain is required. Despite many efforts into TTT, there is a confusion over the experimental settings, thus leading to unfair comparisons. In this work, we first revisit TTT assumptions and categorize TTT protocols by two key factors. Among the multiple protocols, we adopt a realistic sequential test-time training (sTTT) protocol, under which we further develop a test-time anchored clustering (TTAC) approach to enable stronger test-time feature learning. TTAC discovers clusters in both source and target domain and match the target clusters to the source ones to improve generalization. Pseudo label filtering and iterative updating are developed to improve the effectiveness and efficiency of anchored clustering. We demonstrate that under all TTT protocols TTAC consistently outperforms the state-of-the-art methods on six TTT datasets. We hope this work will provide a fair benchmarking of TTT methods and future research should be compared within respective protocols. A demo code is available at https://github.com/Gorilla-Lab-SCUT/TTAC.

Reconstruction of object or scene surfaces has tremendous applications in computer vision, computer graphics, and robotics. The topic attracts increased attention with the emerging pipeline of deep learning surface reconstruction, where implicit field functions constructed from deep networks (e.g., multi-layer perceptrons or MLPs) are proposed for generative shape modeling. In this paper, we study a fundamental problem in this context about recovering a surface mesh from an implicit field function whose zero-level set captures the underlying surface. To achieve the goal, existing methods rely on traditional meshing algorithms (e.g., the de-facto standard marching cubes); while promising, they suffer from loss of precision learned in the implicit surface networks, due to the use of discrete space sampling in marching cubes. Given that an MLP with activations of Rectified Linear Unit (ReLU) partitions its input space into a number of linear regions, we are motivated to connect this local linearity with a same property owned by the desired result of polygon mesh. More specifically, we identify from the linear regions, partitioned by an MLP based implicit function, theanalytic cellsandanalytic facesthat are associated with the function's zero-level isosurface. We prove that under mild conditions, the identified analytic faces are guaranteed to connect and form aclosed, piecewise planar surface. Based on the theorem, we propose an algorithm ofanalytic marching, which marches among analytic cells toexactlyrecover the mesh captured by an implicit surface network. We also show that our theory and algorithm are equally applicable to advanced MLPs with shortcut connections and max pooling. Given the parallel nature of analytic marching, we contributeAnalyticMesh, a software package that supports efficient meshing of implicit surface networks via CUDA parallel computing, and mesh simplification for efficient downstream processing. We apply our method to different settings of generative shape modeling using implicit surface networks. Extensive experiments demonstrate our advantages over existing methods in terms of both meshing accuracy and efficiency. Codes are athttps://github.com/Karbo123/AnalyticMesh.

This paper focuses on the challenging task of learning 3D object surface reconstructions from RGB images. Existing methods achieve varying degrees of success by using different surface representations. However, they all have their own drawbacks, and cannot properly reconstruct the surface shapes of complex topologies, arguably due to a lack of constraints on the topological structures in their learning frameworks. To this end, we propose to learn and use the topology-preserved, skeletal shape representation to assist the downstream task of object surface reconstruction from RGB images. Technically, we propose the novel SkeletonNet design that learns a volumetric representation of a skeleton via a bridged learning of a skeletal point set, where we use parallel decoders each responsible for the learning of points on 1D skeletal curves and 2D skeletal sheets, as well as an efficient module of globally guided subvolume synthesis for a refined, high-resolution skeletal volume; we present a differentiable Point2Voxel layer to make SkeletonNet end-to-end and trainable. With the learned skeletal volumes, we propose two models, the Skeleton-Based Graph Convolutional Neural Network (SkeGCNN) and the Skeleton-Regularized Deep Implicit Surface Network (SkeDISN), which respectively build upon and improve over the existing frameworks of explicit mesh deformation and implicit field learning for the downstream surface reconstruction task. We conduct thorough experiments that verify the efficacy of our proposed SkeletonNet. SkeGCNN and SkeDISN outperform existing methods as well, and they have their own merits when measured by different metrics. Additional results in generalized task settings further demonstrate the usefulness of our proposed methods. We have made our implementation code publicly available at https://github.com/tangjiapeng/SkeletonNet.

Unsupervised domain adaptation (UDA) is to learn classification models that make predictions for unlabeled data on a target domain, given labeled data on a source domain whose distribution diverges from the target one. Mainstream UDA methods strive to learn domain-aligned features such that classifiers trained on the source features can be readily applied to the target ones. Although impressive results have been achieved, these methods have a potential risk of damaging the intrinsic data structures of target discrimination, raising an issue of generalization particularly for UDA tasks in an inductive setting. To address this issue, we are motivated by a UDA assumption of structural similarity across domains, and propose to directly uncover the intrinsic target discrimination via constrained clustering, where we constrain the clustering solutions using structural source regularization that hinges on the very same assumption. Technically, we propose a hybrid model of Structurally Regularized Deep Clustering, which integrates the regularized discriminative clustering of target data with a generative one, and we thus term our method as H-SRDC. Our hybrid model is based on a deep clustering framework that minimizes the Kullback-Leibler divergence between the distribution of network prediction and an auxiliary one, where we impose structural regularization by learning domain-shared classifier and cluster centroids. By enriching the structural similarity assumption, we are able to extend H-SRDC for a pixel-level UDA task of semantic segmentation. We conduct extensive experiments on seven UDA benchmarks of image classification and semantic segmentation. With no explicit feature alignment, our proposed H-SRDC outperforms all the existing methods under both the inductive and transductive settings. We make our implementation codes publicly available at https://github.com/huitangtang/H-SRDC.

Machine learning models have been shown to be vulnerable to adversarial examples. While most of the existing methods for adversarial attack and defense work on the 2D image domain, a few recent attempts have been made to extend them to 3D point cloud data. However, adversarial results obtained by these methods typically contain point outliers, which are both noticeable and easy to defend against using the simple techniques of outlier removal. Motivated by the different mechanisms by which humans perceive 2D images and 3D shapes, in this paper we propose the new design ofgeometry-aware objectives, whose solutions favor (the discrete versions of) the desired surface properties of smoothness and fairness. To generate adversarial point clouds, we use a targeted attack misclassification loss that supports continuous pursuit of increasingly malicious signals. Regularizing the targeted attack loss with our proposed geometry-aware objectives results in our proposed method,Geometry-Aware Adversarial Attack ($GeoA^3$GeoA3). The results of$GeoA^3$tend to be more harmful, arguably harder to defend against, and of the key adversarial characterization of being imperceptible to humans. While the main focus of this paper is to learn to generate adversarial point clouds, we also present a simple but effective algorithm termed$Geo_{+}A^3$-IterNormPro, with Iterative Normal Projection (IterNorPro) that solves a new objective function$Geo_{+}A^3$, towards surface-level adversarial attacks via generation of adversarial point clouds. We quantitatively evaluate our methods on both synthetic and physical objects in terms of attack success rate and geometric regularity. For a qualitative evaluation, we conduct subjective studies by collecting human preferences from Amazon Mechanical Turk. Comparative results in comprehensive experiments confirm the advantages of our proposed methods. Our source codes are publicly available athttps://github.com/Yuxin-Wen/GeoA3.

In this paper, we study the formalism of unsupervised multi-class domain adaptation (multi-class UDA), which underlies a few recent algorithms whose learning objectives are only motivated empirically. Multi-Class Scoring Disagreement (MCSD) divergence is presented by aggregating the absolute margin violations in multi-class classification, and this proposed MCSD is able to fully characterize the relations between any pair of multi-class scoring hypotheses. By using MCSD as a measure of domain distance, we develop a new domain adaptation bound for multi-class UDA; its data-dependent, probably approximately correct bound is also developed that naturally suggests adversarial learning objectives to align conditional feature distributions across source and target domains. Consequently, an algorithmic framework of Multi-class Domain-adversarial learning Networks (McDalNets) is developed, and its different instantiations via surrogate learning objectives either coincide with or resemble a few recently popular methods, thus (partially) underscoring their practical effectiveness. Based on our identical theory for multi-class UDA, we also introduce a new algorithm of Domain-Symmetric Networks (SymmNets), which is featured by a novel adversarial strategy of domain confusion and discrimination. SymmNets affords simple extensions that work equally well under the problem settings of either closed set, partial, or open set UDA. We conduct careful empirical studies to compare different algorithms of McDalNets and our newly introduced SymmNets. Experiments verify our theoretical analysis and show the efficacy of our proposed SymmNets. In addition, we have made our implementation code publicly available.

Estimating 3D human pose from a single image is a challenging task. This work attempts to address the uncertainty of lifting the detected 2D joints to the 3D space by introducing an intermediate state - Part-Centric Heatmap Triplets (HEMlets), which shortens the gap between the 2D observation and the 3D interpretation. The HEMlets utilize three joint-heatmaps to represent the relative depth information of the end-joints for each skeletal body part. In our approach, a Convolutional Network (ConvNet) is first trained to predict HEMlets from the input image, followed by a volumetric joint-heatmap regression. We leverage on the integral operation to extract the joint locations from the volumetric heatmaps, guaranteeing end-to-end learning. Despite the simplicity of the network design, the quantitative comparisons show a significant performance improvement over the best-of-grade methods (e.g., 20 percent on Human3.6M). The proposed method naturally supports training with "in-the-wild" images, where only weakly-annotated relative depth information of skeletal joints is available. This further improves the generalization ability of our model, as validated by qualitative comparisons on outdoor images. Leveraging the strength of the HEMlets pose estimation, we further design and append a shallow yet effective network module to regress the SMPL parameters of the body pose and shape. We term the entire HEMlets-based human pose and shape recovery pipeline HEMlets PoSh. Extensive quantitative and qualitative experiments on the existing human body recovery benchmarks justify the state-of-the-art results obtained with our HEMlets PoSh approach.

Existing deep learning algorithms for point cloud analysis mainly concern discovering semantic patterns from the global configuration of local geometries in a supervised learning manner. However, very few explore geometric properties revealing local surface manifolds embedded in 3-D Euclidean space to discriminate semantic classes or object parts as additional supervision signals. This article is the first attempt to propose a unique multitask geometric learning network to improve semantic analysis by auxiliary geometric learning with local shape properties, which can be either generated via physical computation from point clouds themselves as self-supervision signals or provided as privileged information. Owing to explicitly encoding local shape manifolds in favor of semantic analysis, the proposed geometric self-supervised and privileged learning algorithms can achieve superior performance to their backbone baselines and other state-of-the-art methods, which are verified in the experiments on the popular benchmarks.

Fine-grained visual classification can be addressed by deep representation learning under supervision of manually pre-defined targets (e.g., one-hot or the Hadamard codes). Such target coding schemes are less flexible to model inter-class correlation and are sensitive to sparse and imbalanced data distribution as well. In light of this, this paper introduces a novel target coding scheme - dynamic target relation graphs (DTRG), which, as an auxiliary feature regularization, is a self-generated structural output to be mapped from input images. Specifically, online computation of class-level feature centers is designed to generate cross-category distance in the representation space, which can thus be depicted by a dynamic graph in a non-parametric manner. Explicitly minimizing intra-class feature variations anchored on those class-level centers can encourage learning of discriminative features. Moreover, owing to exploiting inter-class dependency, the proposed target graphs can alleviate data sparsity and imbalanceness in representation learning. Inspired by recent success of the mixup style data augmentation, this paper introduces randomness into soft construction of dynamic target relation graphs to further explore relation diversity of target classes. Experimental results can demonstrate the effectiveness of our method on a number of diverse benchmarks of multiple visual classification, especially achieving the state-of-the-art performance on three popular fine-grained object benchmarks and superior robustness against sparse and imbalanced data. Source codes are made publicly available at https://github.com/AkonLau/DTRG.

The ability to understand the ways to interact with objects from visual cues, a.k.a. visual affordance, is essential to vision-guided robotic research. This involves categorizing, segmenting and reasoning of visual affordance. Relevant studies in 2D and 2.5D image domains have been made previously, however, a truly functional understanding of object affordance requires learning and prediction in the 3D physical domain, which is still absent in the community. In this work, we present a 3D AffordanceNet dataset, a bench-mark of 23k shapes from 23 semantic object categories, annotated with 18 visual affordance categories. Based on this dataset, we provide three benchmarking tasks for evaluating visual affordance understanding, including full-shape, partial-view and rotation-invariant affordance estimations. Three state-of-the-art point cloud deep learning networks are evaluated on all tasks. In addition we also investigate a semi-supervised learning setup to explore the possibility to benefit from unlabeled data. Comprehensive results on our contributed dataset show the promise of visual affordance understanding as a valuable yet challenging benchmark.

This paper focuses on the task of 4D shape reconstruction from a sequence of point clouds. Despite the recent success achieved by extending deep implicit representations into 4D space [29], it is still a great challenge in two respects, i.e. how to design a flexible framework for learning robust spatio-temporal shape representations from 4D point clouds, and develop an efficient mechanism for capturing shape dynamics. In this work, we present a novel pipeline to learn a temporal evolution of the 3D human shape through spatially continuous transformation functions among cross-frame occupancy fields. The key idea is to parallelly establish the dense correspondence between predicted occupancy fields at different time steps via explicitly learning continuous displacement vector fields from robust spatio-temporal shape representations. Extensive comparisons against previous state-of-the-arts show the superior accuracy of our approach for 4D human reconstruction in the problems of 4D shape auto-encoding and completion, and a much faster network inference with about 8 times speedup demonstrates the significant efficiency of our approach. The trained models and implementation code are available at https://github.com/tangjiapeng/LPDC-Net.

Many learning-based approaches have difficulty scaling to unseen data, as the generality of its learned prior is limited to the scale and variations of the training samples. This holds particularly true with 3D learning tasks, given the sparsity of 3D datasets available. We introduce a new learning framework for 3D modeling and reconstruction that greatly improves the generalization ability of a deep generator. Our approach strives to connect the good ends of both learning-based and optimization-based methods. In particular, unlike the common practice that fixes the pre-trained priors at test time, we propose to further optimize the learned prior and latent code according to the input physical measurements after the training. We show that the proposed strategy effectively breaks the barriers constrained by the pre-trained priors and could lead to high-quality adaptation to unseen data. We realize our framework using the implicit surface representation and validate the efficacy of our approach in a variety of challenging tasks that take highly sparse or collapsed observations as input. Experimental results show that our approach compares favorably with the state-of-the-art methods in terms of both generality and accuracy.

Shape modeling and reconstruction from raw point clouds of objects stand as a fundamental challenge in vision and graphics research. Classical methods consider analytic shape priors; however, their performance is degraded when the scanned points deviate from the ideal conditions of cleanness and completeness. Important progress has been recently made by data-driven approaches, which learn global and/or local models of implicit surface representations from auxiliary sets of training shapes. Motivated from a universal phenomenon that self-similar shape patterns of local surface patches repeat across the entire surface of an object, we aim to push forward the data-driven strategies and propose to learn a local implicit surface network for a shared, adaptive modeling of the entire surface for a direct surface reconstruction from raw point cloud; we also enhance the leveraging of surface self-similarities by improving correlations among the optimized latent codes of individual surface patches. Given that orientations of raw points could be unavailable or noisy, we extend signagnostic learning into our local implicit model, which enables our recovery of signed implicit fields of local surfaces from the unsigned inputs. We term our framework as Sign-Agnostic Implicit Learning of Surface Self-Similarities (SAIL-S3). With a global post-optimization of local sign flipping, SAIL-S3 is able to directly model raw, un-oriented point clouds and reconstruct high-quality object surfaces. Experiments show its superiority over existing methods.

Instance segmentation in 3D scenes is fundamental in many applications of scene understanding. It is yet challenging due to the compound factors of data irregularity and uncertainty in the numbers of instances. State-of-the-art methods largely rely on a general pipeline that first learns point-wise features discriminative at semantic and instance levels, followed by a separate step of point grouping for proposing object instances. While promising, they have the shortcomings that (1) the second step is not supervised by the main objective of instance segmentation, and (2) their point-wise feature learning and grouping are less effective to deal with data irregularities, possibly resulting in fragmented segmentations. To address these issues, we propose in this work an end-to-end solution of Semantic Superpoint Tree Network (SSTNet) for proposing object instances from scene points. Key in SST-Net is an intermediate, semantic superpoint tree (SST), which is constructed based on the learned semantic features of superpoints, and which will be traversed and split at intermediate tree nodes for proposals of object instances. We also design in SSTNet a refinement module, termed CliqueNet, to prune superpoints that may be wrongly grouped into instance proposals. Experiments on the benchmarks of ScanNet and S3DIS show the efficacy of our proposed method. At the time of submission, SSTNet ranks top on the ScanNet (V2) leaderboard, with 2% higher of mAP than the second best method. The source code in PyTorch is available at https://github.com/Gorilla-Lab-SCUT/SSTNet.

Category-level 6D object pose and size estimation is to predict full pose configurations of rotation, translation, and size for object instances observed in single, arbitrary views of cluttered scenes. In this paper, we propose a new method of Dual Pose Network with refined learning of pose consistency for this task, shortened as DualPoseNet. DualPoseNet stacks two parallel pose decoders on top of a shared pose encoder, where the implicit decoder predicts object poses with a working mechanism different from that of the explicit one; they thus impose complementary supervision on the training of pose encoder. We construct the encoder based on spherical convolutions, and design a module of Spherical Fusion wherein for a better embedding of pose-sensitive features from the appearance and shape observations. Given no testing CAD models, it is the novel introduction of the implicit decoder that enables the refined pose prediction during testing, by enforcing the predicted pose consistency between the two decoders using a self-adaptive loss term. Thorough experiments on benchmarks of both category- and instance-level object pose datasets confirm efficacy of our designs. DualPoseNet outperforms existing methods with a large margin in the regime of high precision. Our code is released publicly at https://github.com/Gorilla-Lab-SCUT/DualPoseNet.

Surface reconstruction from point clouds is a fundamental problem in the computer vision and graphics community. Recent state-of-the-arts solve this problem by individually optimizing each local implicit field during inference. Without considering the geometric relationships between local fields, they typically require accurate normals to avoid the sign conflict problem in overlapped regions of local fields, which severely limits their applicability to raw scans where surface normals could be unavailable. Although SAL breaks this limitation via sign-agnostic learning, further works still need to explore how to extend this technique for local shape modeling. To this end, we propose to learn implicit surface reconstruction by sign-agnostic optimization of convolutional occupancy networks, to simultaneously achieve advanced scalability to large-scale scenes, generality to novel shapes, and applicability to raw scans in a unified framework. Concretely, we achieve this goal by a simple yet effective design, which further optimizes the pre-trained occupancy prediction networks with an unsigned cross-entropy loss during inference. The learning of occupancy fields is conditioned on convolutional features from an hourglass network architecture. Extensive experimental comparisons with previous state-of-the-arts on both object-level and scene-level datasets demonstrate the superior accuracy of our approach for surface reconstruction from un-orientated point clouds. The code is available at https://github.com/tangjiapeng/SA-ConvONet.

3D LiDAR (light detection and ranging) semantic segmentation is important in scene understanding for many applications, such as auto-driving and robotics. For example, for autonomous cars equipped with RGB cameras and LiDAR, it is crucial to fuse complementary information from different sensors for robust and accurate segmentation. Existing fusion-based methods, however, may not achieve promising performance due to the vast difference between the two modalities. In this work, we investigate a collaborative fusion scheme called perception-aware multi-sensor fusion (PMF) to exploit perceptual information from two modalities, namely, appearance information from RGB images and spatio-depth information from point clouds. To this end, we first project point clouds to the camera coordinates to provide spatio-depth information for RGB images. Then, we propose a two-stream network to extract features from the two modalities, separately, and fuse the features by effective residual-based fusion modules. Moreover, we propose additional perception-aware losses to measure the perceptual difference between the two modalities. Extensive experiments on two benchmark data sets show the superiority of our method. For example, on nuScenes, our PMF outperforms the state-of-the-art method by 0.8% in mIoU.

The point cloud representation of an object can have a large geometric variation in view of inconsistent data acquisition procedure, which thus leads to domain discrepancy due to diverse and uncontrollable shape representation cross datasets. To improve discrimination on unseen distribution of point-based geometries in a practical and feasible perspective, this paper proposes a new method of geometry-aware self-training (GAST) for unsupervised domain adaptation of object point cloud classification. Specifically, this paper aims to learn a domain-shared representation of semantic categories, via two novel self-supervised geometric learning tasks as feature regularization. On one hand, the representation learning is empowered by a linear mixup of point cloud samples with their self-generated rotation labels, to capture a global topological configuration of local geometries. On the other hand, a diverse point distribution across datasets can be normalized with a novel curvature-aware distortion localization. Experiments on the PointDA-10 dataset show that our GAST method can significantly outperform the state-of-the-art methods. Source codes and pre-trained models are available at https://github.com/zou-longkun/GAST.

Existing point cloud classifiers concern on handling irregular data structures to discover a global and discriminative configuration of local geometries. These classification methods design a number of effective permutation-invariant feature encoding kernels, but still suffer from the intrinsic challenge of large geometric feature variations caused by inconsistent point distributions along object surface. In this paper, point cloud classification can be addressed via deep graph representation learning on aggregating multiple convolutional feature kernels (namely, a poly convolutional operation) anchored on each point with its local neighbours. Inspired by recent success of neural architecture search, we introduce a novel concept of poly-convolutional architecture search (PolyConv search in short) to model local geometric patterns in a more flexible manner.

As a basic component of SE(3)-equivariant deep feature learning, steerable convolution has recently demonstrated its advantages for 3D semantic analysis. The advantages are, however, brought by expensive computations on dense, volumetric data, which prevent its practical use for efficient processing of 3D data that are inherently sparse. In this paper, we propose a novel design of Sparse Steerable Convolution (SS-Conv) to address the shortcoming; SS-Conv greatly accelerates steerable convolution with sparse tensors, while strictly preserving the property of SE(3)-equivariance. Based on SS-Conv, we propose a general pipeline for precise estimation of object poses, wherein a key design is a Feature-Steering module that takes the full advantage of SE(3)-equivariance and is able to conduct an efficient pose refinement. To verify our designs, we conduct thorough experiments on three tasks of 3D object semantic analysis, including instance-level 6D pose estimation, category-level 6D pose and size estimation, and category-level 6D pose tracking. Our proposed pipeline based on SS-Conv outperforms existing methods on almost all the metrics evaluated by the three tasks. Ablation studies also show the superiority of our SS-Conv over alternative convolutions in terms of both accuracy and efficiency. Our code is released publicly at https://github.com/Gorilla-Lab-SCUT/SS-Conv.

Skeleton-based action recognition has attracted research attentions in recent years. One common drawback in currently popular skeleton-based human action recognition methods is that the sparse skeleton information alone is not sufficient to fully characterize human motion. This limitation makes several existing methods incapable of correctly classifying action categories which exhibit only subtle motion differences. In this paper, we propose a novel framework for employing human pose skeleton and joint-centered light-weight information jointly in a two-stream graph convolutional network, namely, JOLO-GCN. Specifically, we use Joint-aligned optical Flow Patches (JFP) to capture the local subtle motion around each joint as the pivotal joint-centered visual information. Compared to the pure skeleton-based baseline, this hybrid scheme effectively boosts performance, while keeping the computational and memory overheads low. Experiments on the NTU RGB+D, NTU RGB+D 120, and the Kinetics-Skeleton dataset demonstrate clear accuracy improvements attained by the proposed method over the state-of-the-art skeleton-based methods.

In this paper, we introduce the algorithms of Orthogonal Deep Neural Networks (OrthDNNs) to connect with recent interest of spectrally regularized deep learning methods. OrthDNNs are theoretically motivated by generalization analysis of modern DNNs, with the aim to find solution properties of network weights that guarantee better generalization. To this end, we first prove that DNNs are of local isometry on data distributions of practical interest; by using a new covering of the sample space and introducing the local isometry property of DNNs into generalization analysis, we establish a new generalization error bound that is both scale- and range-sensitive to singular value spectrum of each of networks' weight matrices. We prove that the optimal bound w.r.t. the degree of isometry is attained when each weight matrix has a spectrum of equal singular values, among which orthogonal weight matrix or a non-square one with orthonormal rows or columns is the most straightforward choice, suggesting the algorithms of OrthDNNs. We present both algorithms of strict and approximate OrthDNNs, and for the later ones we propose a simple yet effective algorithm called Singular Value Bounding (SVB), which performs as well as strict OrthDNNs, but at a much lower computational cost. We also propose Bounded Batch Normalization (BBN) to make compatible use of batch normalization with OrthDNNs. We conduct extensive comparative studies by using modern architectures on benchmark image classification. Experiments show the efficacy of OrthDNNs.

Given labeled instances on a source domain and unlabeled ones on a target domain, unsupervised domain adaptation aims to learn a task classifier that can well classify target instances. Recent advances rely on domain-adversarial training of deep networks to learn domain-invariant features. However, due to an issue of mode collapse induced by the separate design of task and domain classifiers, these methods are limited in aligning the joint distributions of feature and category across domains. To overcome it, we propose a novel adversarial learning method termed Discriminative Adversarial Domain Adaptation (DADA). Based on an integrated category and domain classifier, DADA has a novel adversarial objective that encourages a mutually inhibitory relation between category and domain predictions for any input instance. We show that under practical conditions, it defines a minimax game that can promote the joint distribution alignment. Except for the traditional closed set domain adaptation, we also extend DADA for extremely challenging problem settings of partial and open set domain adaptation. Experiments show the efficacy of our proposed methods and we achieve the new state of the art for all the three settings on benchmark datasets.

Regressing the illumination of a scene from the representations of object appearances is popularly adopted in computational color constancy. However, it's still challenging due to intrinsic appearance and label ambiguities caused by unknown illuminants, diverse reflection properties of materials and extrinsic imaging factors (such as different camera sensors). In this paper, we introduce a novel algorithm – Cascading Convolutional Color Constancy (in short, C4) to improve robustness of regression learning and achieve stable generalization capability across datasets (different cameras and scenes) in a unique framework. The proposed C4 method ensembles a series of dependent illumination hypotheses from each cascade stage via introducing a weighted multiply-accumulate loss function, which can inherently capture different modes of illuminations and explicitly enforce coarse-to-fine network optimization. Experimental results on the public Color Checker and NUS 8-Camera benchmarks demonstrate superior performance of the proposed algorithm in comparison with the state-of-the-art methods, especially for more difficult scenes.

We propose to incorporate neural architecture search (NAS) into general-purpose multi-task learning (GP-MTL). Existing NAS methods typically define different search spaces according to different tasks. In order to adapt to different task combinations (i.e., task sets), we disentangle the GP-MTL networks into single-task backbones (optionally encode the task priors), and a hierarchical and layerwise features sharing/fusing scheme across them. This enables us to design a novel and general task-agnostic search space, which inserts cross-task edges (i.e., feature fusion connections) into fixed single-task network backbones. Moreover, we also propose a novel single-shot gradient-based search algorithm that closes the performance gap between the searched architectures and the final evaluation architecture. This is realized with a minimum entropy regularization on the architecture weights during the search phase, which makes the architecture weights converge to near-discrete values and therefore achieves a single model. As a result, our searched model can be directly used for evaluation without (re-)training from scratch. We perform extensive experiments using different single-task backbones on various task sets, demonstrating the promising performance obtained by exploiting the hierarchical and layerwise features, as well as the desirable generalizability to different i) task sets and ii) single-task backbones. The code of our paper is available at https://github.com/bhpfelix/MTLNAS.

Unsupervised domain adaptation (UDA) is to make predictions for unlabeled data on a target domain, given labeled data on a source domain whose distribution shifts from the target one. Mainstream UDA methods learn aligned features between the two domains, such that a classifier trained on the source features can be readily applied to the target ones. However, such a transferring strategy has a potential risk of damaging the intrinsic discrimination of target data. To alleviate this risk, we are motivated by the assumption of structural domain similarity, and propose to directly uncover the intrinsic target discrimination via discriminative clustering of target data. We constrain the clustering solutions using structural source regularization that hinges on our assumed structural domain similarity. Technically, we use a flexible framework of deep network based discriminative clustering that minimizes the KL divergence between predictive label distribution of the network and an introduced auxiliary one; replacing the auxiliary distribution with that formed by ground-truth labels of source data implements the structural source regularization via a simple strategy of joint network training. We term our proposed method as Structurally Regularized Deep Clustering (SRDC), where we also enhance target discrimination with clustering of intermediate network features, and enhance structural regularization with soft selection of less divergent source examples. Careful ablation studies show the efficacy of our proposed SRDC. Notably, with no explicit domain alignment, SRDC outperforms all existing methods on three UDA benchmarks.

This paper studies a problem of learning surface mesh via implicit functions in an emerging field of deep learning surface reconstruction, where implicit functions are popularly implemented as multi-layer perceptrons (MLPs) with rectified linear units (ReLU). To achieve meshing from the learned implicit functions, existing methods adopt the de-facto standard algorithm of marching cubes; while promising, they suffer from loss of precision learned in the MLPs, due to the discretization nature of marching cubes. Motivated by the knowledge that a ReLU based MLP partitions its input space into a number of linear regions, we identify from these regions analytic cells and faces that are associated with zero-level isosurface of the implicit function, and characterize the conditions under which the identified faces are guaranteed to connect and form a closed, piecewise planar surface. We propose a naturally parallelizable algorithm of analytic marching to exactly recover the mesh captured by a learned MLP. Experiments on deep learning mesh reconstruction verify the advantages of our algorithm over existing ones.

The problem of adversarial examples has shown that modern Neural Network (NN) models could be rather fragile. Among the more established techniques to solve the problem, one is to require the model to be \emph{$\epsilon$-adversarially robust} (AR); that is, to require the model not to change predicted labels when any given input examples are perturbed within a certain range. However, it is observed that such methods would lead to standard performance degradation, i.e., the degradation on natural examples. In this work, we study the degradation through the regularization perspective. We identify quantities from generalization analysis of NNs; with the identified quantities we empirically find that AR is achieved by regularizing/biasing NNs towards less confident solutions by making the changes in the feature space (induced by changes in the instance space) of most layers smoother uniformly in all directions; so to a certain extent, it prevents sudden change in prediction w.r.t. perturbations. However, the end result of such smoothing concentrates samples around decision boundaries, resulting in less confident solutions, and leads to worse standard performance. Our studies suggest that one might consider ways that build AR into NNs in a gentler way to avoid the problematic regularization.

Learning robotic grasps from visual observations is a promising yet challenging task. Recent research shows its great potential by preparing and learning from large-scale synthetic datasets. For the popular, 6 degree-of-freedom (6-DOF) grasp setting of parallel-jaw gripper, most of existing methods take the strategy of heuristically sampling grasp candidates and then evaluating them using learned scoring functions. This strategy is limited in terms of the conflict between sampling efficiency and coverage of optimal grasps. To this end, we propose in this work a novel, end-to-end \emph{Grasp Proposal Network (GPNet)}, to predict a diverse set of 6-DOF grasps for an unseen object observed from a single and unknown camera view. GPNet builds on a key design of grasp proposal module that defines \emph{anchors of grasp centers} at discrete but regular 3D grid corners, which is flexible to support either more precise or more diverse grasp predictions. To test GPNet, we contribute a synthetic dataset of 6-DOF object grasps; evaluation is conducted using rule-based criteria, simulation test, and real test. Comparative results show the advantage of our methods over existing ones. Notably, GPNet gains better simulation results via the specified coverage, which helps achieve a ready translation in real test. Our code and dataset are available on \url{https://github.com/CZ-Wu/GPNet}.

Fine-grained object categorization aims for distinguishing objects of subordinate categories that belong to the same entry-level object category. It is a rapidly developing subfield in multimedia content analysis. The task is challenging due to the facts that (1) training images with ground-truth labels are difficult to obtain, and (2) variations among different subordinate categories are subtle. It is well established that characterizing features of different subordinate categories are located on local parts of object instances. However, manually annotating object parts requires expertise, which is also difficult to generalize to new fine-grained categorization tasks. In this work, we propose a Weakly Supervised Part Detection Network (PartNet) that is able to detect discriminative local parts for the use of fine-grained categorization. A vanilla PartNet builds on top of a base subnetwork two parallel streams of upper network layers, which respectively compute scores of classification probabilities (over subordinate categories) and detection probabilities (over a specified number of discriminative part detectors) for local regions of interest (RoIs). The image-level prediction is obtained by aggregating element-wise products of these region-level probabilities, and meanwhile diverse part detectors can be learned in an end-to-end fashion under the image-level supervision. To generate a diverse set of RoIs as inputs of PartNet, we propose a simple Discretized Part Proposals module (DPP) that directly targets for proposing candidates of discriminative local parts, with no bridging via object-level proposals. Experiments on benchmark datasets of CUB-200-2011, Oxford Flower 102 and Oxford-IIIT Pet show the efficacy of our proposed method for both discriminative part detection and fine-grained categorization. In particular, we achieve the new state-of-the-art performance on CUB-200-2011 and Oxford-IIIT Pet datasets when ground-truth part annotations are not available.

This paper focuses on the challenging task of learning 3D object surface reconstructions from single RGB images. Existing methods achieve varying degrees of success by using different geometric representations. However, they all have their own drawbacks, and cannot well reconstruct those surfaces of complex topologies. To this end, we propose in this paper a skeleton-bridged, stage-wise learning approach to address the challenge. Our use of skeleton is due to its nice property of topology preservation, while being of lower complexity to learn. To learn skeleton from an input image, we design a deep architecture whose decoder is based on a novel design of parallel streams respectively for synthesis of curve- and surface-like skeleton points. We use different shape representations of point cloud, volume, and mesh in our stage-wise learning, in order to take their respective advantages. We also propose multi-stage use of the input image to correct prediction errors that are possibly accumulated in each stage. We conduct intensive experiments to investigate the efficacy of our proposed approach. Qualitative and quantitative results on representative object categories of both simple and complex topologies demonstrate the superiority of our approach over existing ones. We will make our ShapeNet-Skeleton dataset publicly available.

Unsupervised domain adaptation aims to learn a model of classifier for unlabeled samples on the target domain, given training data of labeled samples on the source domain. Impressive progress is made recently by learning invariant features via domain-adversarial training of deep networks. In spite of the recent progress, domain adaptation is still limited in achieving the invariance of feature distributions at a finer category level. To this end, we propose in this paper a new domain adaptation method called Domain-Symmetric Networks (SymNets). The proposed SymNet is based on a symmetric design of source and target task classifiers, based on which we also construct an additional classifier that shares with them its layer neurons. To train the SymNet, we propose a novel adversarial learning objective whose key design is based on a two-level domain confusion scheme, where the category-level confusion loss improves over the domain-level one by driving the learning of intermediate network features to be invariant at the corresponding categories of the two domains. Both domain discrimination and domain confusion are implemented based on the constructed additional classifier. Since target samples are unlabeled, we also propose a scheme of cross-domain training to help learn the target classifier. Careful ablation studies show the efficacy of our proposed method. In particular, based on commonly used base networks, our SymNets achieve the new state of the art on three benchmark domain adaptation datasets.

Reconstructing the 3D mesh of a general object from a single image is now possible thanks to the latest advances of deep learning technologies. However, due to the nontrivial difficulty of generating a feasible mesh structure, the state-of-the-art approaches often simplify the problem by learning the displacements of a template mesh that deforms it to the target surface. Though reconstructing a 3D shape with complex topology can be achieved by deforming multiple mesh patches, it remains difficult to stitch the results to ensure a high meshing quality. In this paper, we present an end-to-end single-view mesh reconstruction framework that is able to generate high-quality meshes with complex topologies from a single genus-0 template mesh. The key to our approach is a novel progressive shaping framework that alternates between mesh deformation and topology modification. While a deformation network predicts the per-vertex translations that reduce the gap between the reconstructed mesh and the ground truth, a novel topology modification network is employed to prune the error-prone faces, enabling the evolution of topology. By iterating over the two procedures, one can progressively modify the mesh topology while achieving higher reconstruction accuracy. Moreover, a boundary refinement network is designed to refine the boundary conditions to further improve the visual quality of the reconstructed mesh. Extensive experiments demonstrate that our approach outperforms the current state-of-the-art methods both qualitatively and quantitatively, especially for the shapes with complex topologies.

Few-shot learning presents a challenge that a classifier must quickly adapt to new classes that do not appear in the training set, given only a few labeled examples of each new class. This paper proposes a position-aware relation network (PARN) to learn a more flexible and robust metric ability for few-shot learning. Relation networks (RNs), a kind of architectures for relational reasoning, can acquire a deep metric ability for images by just being designed as a simple convolutional neural network (CNN)[23]. However, due to the inherent local connectivity of CNN, the CNN-based relation network (RN) can be sensitive to the spatial position relationship of semantic objects in two compared images. To address this problem, we introduce a deformable feature extractor (DFE) to extract more efficient features, and design a dual correlation attention mechanism (DCA) to deal with its inherent local connectivity. Successfully, our proposed approach extents the potential of RN to be position-aware of semantic objects by introducing only a small number of parameters. We evaluate our approach on two major benchmark datasets, i.e., Omniglot and Mini-Imagenet, and on both of the datasets our approach achieves state-of-the-art performance. It's worth noting that our 5-way 1-shot result on Omniglot even outperforms the previous 5-way 5-shot results.

Estimating 3D human pose from a single image is a challenging task. This work attempts to address the uncertainty of lifting the detected 2D joints to the 3D space by introducing an intermediate state - Part-Centric Heatmap Triplets (HEMlets), which shortens the gap between the 2D observation and the 3D interpretation. The HEMlets utilize three joint-heatmaps to represent the relative depth information of the end-joints for each skeletal body part. In our approach, a Convolutional Network(ConvNet) is first trained to predict HEMlests from the input image, followed by a volumetric joint-heatmap regression. We leverage on the integral operation to extract the joint locations from the volumetric heatmaps, guaranteeing end-to-end learning. Despite the simplicity of the network design, the quantitative comparisons show a significant performance improvement over the best-of-grade method (by 20% on Human3.6M). The proposed method naturally supports training with "in-the-wild'' images, where only weakly-annotated relative depth information of skeletal joints is available. This further improves the generalization ability of our model, as validated by qualitative comparisons on outdoor images.

This paper proposes a deep cascade network to generate 3D geometry of an object on a point cloud, consisting of a set of permutation-insensitive points. Such a surface representation is easy to learn from, but inhibits exploiting rich low-dimensional topological manifolds of the object shape due to lack of geometric connectivity. For benefiting from its simple structure yet utilizing rich neighborhood information across points, this paper proposes a two-stage cascade model on point sets. Specifically, our method adopts the state-of-the-art point set autoencoder to generate a sparsely coarse shape first, and then locally refines it by encoding neighborhood connectivity on a graph representation. An ensemble of sparse refined surface is designed to alleviate the suffering from local minima caused by modeling complex geometric manifolds. Moreover, our model develops a dynamically-weighted loss function for jointly penalizing the generation output of cascade levels at different training stages in a coarse-to-fine manner. Comparative evaluation on the publicly benchmarking ShapeNet dataset demonstrates superior performance of the proposed model to the state-of-the-art methods on both single-view shape reconstruction and shape autoencoding applications.

In this work, we propose a novel method termed Frustum ConvNet (F-ConvNet) for amodal 3D object detection from point clouds. Given 2D region proposals in an RGB image, our method first generates a sequence of frustums for each region proposal, and uses the obtained frustums to group local points. F-ConvNet aggregates point-wise features as frustum-level feature vectors, and arrays these feature vectors as a feature map for use of its subsequent component of fully convolutional network (FCN), which spatially fuses frustum-level features and supports an end-to-end and continuous estimation of oriented boxes in the 3D space. We also propose component variants of F-ConvNet, including an FCN variant that extracts multi-resolution frustum features, and a refined use of F-ConvNet over a reduced 3D space. Careful ablation studies verify the efficacy of these component variants. F-ConvNet assumes no prior knowledge of the working 3D environment and is thus dataset-agnostic. We present experiments on both the indoor SUN-RGBD and outdoor KITTI datasets. F-ConvNet outperforms all existing methods on SUN-RGBD, and at the time of submission it outperforms all published works on the KITTI benchmark. Code has been made available at: https://github.com/zhixinwang/frustum-convnet.

Multiple marginal matching problem aims at learning mappings to match a source domain to multiple target domains and it has attracted great attention in many applications, such as multi-domain image translation. However, addressing this problem has two critical challenges: (i) Measuring the multi-marginal distance among different domains is very intractable; (ii) It is very difficult to exploit cross-domain correlations to match the target domain distributions. In this paper, we propose a novel Multi-marginal Wasserstein GAN (MWGAN) to minimize Wasserstein distance among domains. Specifically, with the help of multi-marginal optimal transport theory, we develop a new adversarial objective function with inner- and inter-domain constraints to exploit cross-domain correlations. Moreover, we theoretically analyze the generalization performance of MWGAN, and empirically evaluate it on the balanced and imbalanced translation tasks. Extensive experiments on toy and real-world datasets demonstrate the effectiveness of MWGAN.

Two-stage Cumulative Attribute (CA) regression has been found effective in regression problems of computer vision such as facial age and crowd density estimation. The first stage regression maps input features to cumulative attributes that encode correlations between target values. The previous works have dealt with single output regression. In this work, we propose cumulative attribute spaces for 2- and 3-output (multivariate) regression. We show how the original CA space can be generalized to multiple output by the Cartesian product (CartCA). However, for target spaces with more than two outputs the CartCA becomes computationally infeasible and therefore we propose an approximate solution - multi-view CA (MvCA) - where CartCA is applied to output pairs. We experimentally verify improved performance of the CartCA and MvCA spaces in 2D and 3D face pose estimation and three-output (RGB) illuminant estimation for color constancy.

Objectness has recently become a standard step in many computer vision tasks. Among various techniques, those based on hierarchical image segmentation play a fundamental role for developments in new data modalities. In this paper, we address the problem of objectness in RGB-D images and propose a novel and effective approach, namely, hierarchical image segmentation ensemble (HISE). Different from existing image segmentation based methods that generate object segments or proposals largely by heuristics or empirical rules, HISE learns superpixel mergings with a hierarchical tree-structured ensemble, where individual merging models of the ensemble are formed by traversing different paths of the tree, and where both the merging accuracy and proposal diversity are emphasized. Furthermore, we use efficient feature measurements that support easy integration of additional clues. Extensive experiments conducted on the benchmark NYU-v2 RGB-D and SUN RGB-D data sets show the competency of our proposed method.

Many machine learning problems concern with discovering or associating common patterns in data of multiple views or modalities. Multi-view learning is of the methods to achieve such goals. Recent methods propose deep multi-view networks via adaptation of generic Deep Neural Networks (DNNs), which concatenate features of individual views at intermediate network layers (i.e., fusion layers). In this work, we study the problem of multi-view learning in such end-to-end networks. We take a regularization approach via multi-view learning criteria, and propose a novel, effective, and efficient neuron-wise correlation-maximizing regularizer. We implement our proposed regularizers collectively as a correlation-regularized network layer (CorrReg). CorrReg can be applied to either fully-connected or convolutional fusion layers, simply by replacing them with their CorrReg counterparts. By partitioning neurons of a hidden layer in generic DNNs into multiple subsets, we also consider a multi-view feature learning perspective of generic DNNs. Such a perspective enables us to study deep multi-view learning in the context of regularized network training, for which we present control experiments of benchmark image classification to show the efficacy of our proposed CorrReg. To investigate how CorrReg is useful for practical multi-view learning problems, we conduct experiments of RGB-D object/scene recognition and multi-view based 3D object recognition, using networks with fusion layers that concatenate intermediate features of individual modalities or views for subsequent classification. Applying CorrReg to fusion layers of these networks consistently improves classification performance. In particular, we achieve the new state of the art on the benchmark RGB-D object and RGB-D scene datasets. We make the implementation of CorrReg publicly available.

This work presents a novel architecture of deep neural networks to generate meshes approximating the surface of a 3D object from a single image. Compared to existing learning-based 3D reconstruction models, our architecture is characterized by (1) deep mesh deformation stacks with residual network design, where a simple mesh is transformed to approximate the target surface and undergoes multiple deformation steps to progressively refine the result and reduce the residuals, and (2) parallel paths per deformation step, which can exponentially enrich the generated meshes using deeper structure and more model parameters. We also propose novel regularization scheme that encourages the meshes to be both globally complementary to cover the target surface and locally consistent with each other. Empirical evaluation on benchmark datasets show advantage of the proposed architecture over existing methods.

Deep learning methods achieve great success recently on many computer vision problems. In spite of these practical successes, optimization of deep networks remains an active topic in deep learning research. In this work, we focus on investigation of the network solution properties that can potentially lead to good performance. Our research is inspired by theoretical and empirical results that use orthogonal matrices to initialize networks, but we are interested in investigating how orthogonal weight matrices perform when network training converges. To this end, we propose to constrain the solutions of weight matrices in the orthogonal feasible set during the whole process of network training, and achieve this by a simple yet effective method called Singular Value Bounding (SVB). In SVB, all singular values of each weight matrix are simply bounded in a narrow band around the value of 1. Based on the same motivation, we also propose Bounded Batch Normalization (BBN), which improves Batch Normalization by removing its potential risk of ill-conditioned layer transform. We present both theoretical and empirical results to justify our proposed methods. Experiments on benchmark image classification datasets show the efficacy of our proposed SVB and BBN. In particular, we achieve the state-of-the-art results of 3.06% error rate on CIFAR10 and 16.90% on CIFAR100, using off-the-shelf network architectures (Wide ResNets). Our preliminary results on ImageNet also show the promise in large-scale learning. We release the implementation code of our methods at www.aperture-lab.net/research/svb.

Human actions captured in video sequences are threedimensional signals characterizing visual appearance and motion dynamics. To learn action patterns, existing methods adopt Convolutional and/or Recurrent Neural Networks (CNNs and RNNs). CNN based methods are effective in learning spatial appearances, but are limited in modeling long-term motion dynamics. RNNs, especially Long Short- Term Memory (LSTM), are able to learn temporal motion dynamics. However, naively applying RNNs to video sequences in a convolutional manner implicitly assumes that motions in videos are stationary across different spatial locations. This assumption is valid for short-term motions but invalid when the duration of the motion is long.,,In this work, we propose Lattice-LSTM (L2STM), which extends LSTM by learning independent hidden state transitions of memory cells for individual spatial locations. This method effectively enhances the ability to model dynamics across time and addresses the non-stationary issue of long-term motion dynamics without significantly increasing the model complexity. Additionally, we introduce a novel multi-modal training procedure for training our network. Unlike traditional two-stream architectures which use RGB and optical flow information as input, our two-stream model leverages both modalities to jointly train both input gates and both forget gates in the network rather than treating the two streams as separate entities with no information about the other. We apply this end-to-end system to benchmark datasets (UCF-101 and HMDB-51) of human action recognition. Experiments show that on both datasets, our proposed method outperforms all existing ones that are based on LSTM and/or CNNs of similar model complexities.

This paper presents a patch-set-based sparse representation for image set classification. Compared with image-based image set representation, our patch-set-based representation is alignment free and thus has an advantage for tasks like video-based face recognition, image-set-based object recognition, and video-based hand gesture recognition, where precious alignment is usually difficult or even impossible due to large variance in view angle or pose. Specifically, to bypass the alignment issue, we propose to adopt the patch-based image set representation by dividing each image within each set into patches, then we cluster all the training patches into multiple clusters and classify the test patches based on the cluster centers of training patches. The labels of test patches within each cluster are inferred from a patch-set-based sparse representation for classification, and the labels of all test patches from all the clusters are then aggregated to predict a single label for the test set. Experimental results on video-based face recognition data sets (CMU-MoBo and YouTube Celebrities), image-set-based object recognition data set (ETH-80), and video-based hand gesture recognition data set (Kinect Hand Gestures) demonstrate that our proposed method consistently outperforms all existing ones, and the improvement is very significant on the YouTube Celebrities and Kinect Hand Gesture data sets. Moreover, we also quantitatively show the robustness of our method to misalignment on the Mutli-PIE data set.

Single image haze removal is a challenging ill-posed problem. Existing methods use various constraints/priors to get plausible dehazing solutions. The key to achieve haze removal is to estimate a medium transmission map for an input hazy image. In this paper, we propose a trainable end-to-end system called DehazeNet, for medium transmission estimation. DehazeNet takes a hazy image as input, and outputs its medium transmission map that is subsequently used to recover a haze-free image via atmospheric scattering model. DehazeNet adopts convolutional neural network-based deep architecture, whose layers are specially designed to embody the established assumptions/priors in image dehazing. Specifically, the layers of Maxout units are used for feature extraction, which can generate almost all haze-relevant features. We also propose a novel nonlinear activation function in DehazeNet, called bilateral rectified linear unit, which is able to improve the quality of recovered haze-free image. We establish connections between the components of the proposed DehazeNet and those used in existing methods. Experiments on benchmark images show that DehazeNet achieves superior performance over existing methods, yet keeps efficient and easy to use.

Visual tracking is challenging due to image variations caused by various factors, such as object deformation, scale change, illumination change, and occlusion. Given the superior tracking performance of human visual system (HVS), an ideal design of biologically inspired model is expected to improve computer visual tracking. This is, however, a difficult task due to the incomplete understanding of neurons' working mechanism in the HVS. This paper aims to address this challenge based on the analysis of visual cognitive mechanism of the ventral stream in the visual cortex, which simulates shallow neurons (S1 units and C1 units) to extract low-level biologically inspired features for the target appearance and imitates an advanced learning mechanism (S2 units and C2 units) to combine generative and discriminative models for target location. In addition, fast Gabor approximation and fast Fourier transform are adopted for real-time learning and detection in this framework. Extensive experiments on large-scale benchmark data sets show that the proposed biologically inspired tracker performs favorably against the state-of-the-art methods in terms of efficiency, accuracy, and robustness. The acceleration technique in particular ensures that biologically inspired tracker maintains a speed of approximately 45 frames/s.

Human actions in video sequences are three-dimensional (3D) spatio-temporal signals characterizing both the visual appearance and motion dynamics of the involved humans and objects. Inspired by the success of convolutional neural networks (CNN) for image classification, recent attempts have been made to learn 3D CNNs for recognizing human actions in videos. However, partly due to the high complexity of training 3D convolution kernels and the need for large quantities of training videos, only limited success has been reported. This has triggered us to investigate in this paper a new deep architecture which can handle 3D signals more effectively. Specifically, we propose factorized spatio-temporal convolutional networks (FstCN) that factorize the original 3D convolution kernel learning as a sequential process of learning 2D spatial kernels in the lower layers (called spatial convolutional layers), followed by learning 1D temporal kernels in the upper layers (called temporal convolutional layers). We introduce a novel transformation and permutation operator to make factorization in FstCN possible. Moreover, to address the issue of sequence alignment, we propose an effective training and inference strategy based on sampling multiple video clips from a given action video sequence. We have tested FstCN on two commonly used benchmark datasets (UCF-101 and HMDB-51). Without using auxiliary training videos to boost the performance, FstCN outperforms existing CNN based methods and achieves comparable performance with a recent method that benefits from using auxiliary training videos.

In this paper, we present a subcategory-aware recognition framework to boost category level object classification performance. Different from the existing monolithic model approaches, we aim to automatically leverage the embedded subcategory structure to assist the further category level recognition. Motivated by the observation of considerable intra-class diversities and inter-class ambiguities in many current object classification data sets, we explicitly split data into subcategories by ambiguity-guided subcategory mining. The resulting subcategories are seamlessly integrated into the state-of-the-art detection-assisted classification framework. In particular, we build the instance affinity graph by combining both intra-class similarity and inter-class ambiguity. Visual subcategories, which correspond to the dense subgraphs, are detected by the graph shift algorithm. We then train an individual model for each subcategory rather than an attempt to represent an object category with a monolithic model. Related samples, which are informative for subcategory classification, are utilized to regularize each subcategory model. Finally, the responses from subcategory models are aggregated by subcategory-aware kernel regression. The extensive experiments over the PASCAL visual object challenge (VOC) 2007 and PASCAL VOC 2010 databases show the state-of-the-art performance from our framework.

Gait, as a promising biometric for recognizing human identities, can be nonintrusively captured as a series of acceleration signals using wearable or portable smart devices. It can be used for access control. Most existing methods on accelerometer-based gait recognition require explicit step-cycle detection, suffering from cycle detection failures and intercycle phase misalignment. We propose a novel algorithm that avoids both the above two problems. It makes use of a type of salient points termed signature points (SPs), and has three components: 1) a multiscale SP extraction method, including the localization and SP descriptors; 2) a sparse representation scheme for encoding newly emerged SPs with known ones in terms of their descriptors, where the phase propinquity of the SPs in a cluster is leveraged to ensure the physical meaningfulness of the codes; and 3) a classifier for the sparse-code collections associated with the SPs of a series. Experimental results on our publicly available dataset of 175 subjects showed that our algorithm outperformed existing methods, even if the step cycles were perfectly detected for them. When the accelerometers at five different body locations were used together, it achieved the rank-1 accuracy of 95.8% for identification, and the equal error rate of 2.2% for verification.

This paper targets learning robust image representation for single training sample per person face recognition. Motivated by the success of deep learning in image representation, we propose a supervised autoencoder, which is a new type of building block for deep architectures. There are two features distinct our supervised autoencoder from standard autoencoder. First, we enforce the faces with variants to be mapped with the canonical face of the person, for example, frontal face with neutral expression and normal illumination; Second, we enforce features corresponding to the same person to be similar. As a result, our supervised autoencoder extracts the features which are robust to variances in illumination, expression, occlusion, and pose, and facilitates the face recognition. We stack such supervised autoencoders to get the deep architecture and use it for extracting features in image representation. Experimental results on the AR, Extended Yale B, CMU-PIE, and Multi-PIE data sets demonstrate that by coupling with the commonly used sparse representation-based classification, our stacked supervised autoencoders-based face representation significantly outperforms the commonly used image representations in single sample per person face recognition, and it achieves higher recognition accuracy compared with other deep learning models, including the deep Lambertian network, in spite of much less training data and without any domain information. Moreover, supervised autoencoder can also be used for face verification, which further demonstrates its effectiveness for face representation.

In this paper, we propose a reconstruction-based metric learning method to learn a discriminative distance metric for unconstrained face verification. Unlike conventional metric learning methods, which only consider the label information of training samples and ignore the reconstruction residual information in the learning procedure, we apply a reconstruction criterion to learn a discriminative distance metric. For each training example, the distance metric is learned by enforcing a margin between the interclass sparse reconstruction residual and interclass sparse reconstruction residual, so that the reconstruction residual of training samples can be effectively exploited to compute the between-class and within-class variations. To better use multiple features for distance metric learning, we propose a reconstruction-based multimetric learning method to collaboratively learn multiple distance metrics, one for each feature descriptor, to remove uncorrelated information for recognition. We evaluate our proposed methods on the Labelled Faces in the Wild (LFW) and YouTube face data sets and our experimental results clearly show the superiority of our methods over both previous metric learning methods and several state-of-the-art unconstrained face verification methods.

In this paper, we propose a very simple deep learning network for image classification that is based on very basic data processing components: 1) cascaded principal component analysis (PCA); 2) binary hashing; and 3) blockwise histograms. In the proposed architecture, the PCA is employed to learn multistage filter banks. This is followed by simple binary hashing and block histograms for indexing and pooling. This architecture is thus called the PCA network (PCANet) and can be extremely easily and efficiently designed and learned. For comparison and to provide a better understanding, we also introduce and study two simple variations of PCANet: 1) RandNet and 2) LDANet. They share the same topology as PCANet, but their cascaded filters are either randomly selected or learned from linear discriminant analysis. We have extensively tested these basic networks on many benchmark visual data sets for different tasks, including Labeled Faces in the Wild (LFW) for face verification; the MultiPIE, Extended Yale B, AR, Facial Recognition Technology (FERET) data sets for face recognition; and MNIST for hand-written digit recognition. Surprisingly, for all tasks, such a seemingly naive PCANet model is on par with the state-of-the-art features either prefixed, highly hand-crafted, or carefully learned [by deep neural networks (DNNs)]. Even more surprisingly, the model sets new records for many classification tasks on the Extended Yale B, AR, and FERET data sets and on MNIST variations. Additional experiments on other public data sets also demonstrate the potential of PCANet to serve as a simple but highly competitive baseline for texture classification and object recognition.

This paper proposes a joint multi-task learning algorithm to better predict attributes in images using deep convolutional neural networks (CNN). We consider learning binary semantic attributes through a multi-task CNN model, where each CNN will predict one binary attribute. The multi-task learning allows CNN models to simultaneously share visual knowledge among different attribute categories. Each CNN will generate attribute-specific feature representations, and then we apply multi-task learning on the features to predict their attributes. In our multi-task framework, we propose a method to decompose the overall model's parameters into a latent task matrix and combination matrix. Furthermore, under-sampled classifiers can leverage shared statistics from other classifiers to improve their performance. Natural grouping of attributes is applied such that attributes in the same group are encouraged to share more knowledge. Meanwhile, attributes in different groups will generally compete with each other, and consequently share less knowledge. We show the effectiveness of our method on two popular attribute datasets.

Most of the previous work on video action recognition use complex hand-designed local features, such as SIFT, HOG and SURF, but these approaches are implemented sophisticatedly and difficult to be extended to other sensor modalities. Recent studies discover that there are no universally best hand-engineered features for all datasets, and learning features directly from the data may be more advantageous. One such endeavor is Slow Feature Analysis (SFA) proposed by Wiskott and Sejnowski [33]. SFA can learn the invariant and slowly varying features from input signals and has been proved to be valuable in human action recognition [34]. It is also observed that the multi-layer feature representation has succeeded remarkably in widespread machine learning applications. In this paper, we propose to combine SFA with deep learning techniques to learn hierarchical representations from the video data itself. Specifically, we use a two-layered SFA learning structure with 3D convolution and max pooling operations to scale up the method to large inputs and capture abstract and structural features from the video. Thus, the proposed method is suitable for action recognition. At the same time, sharing the same merits of deep learning, the proposed method is generic and fully automated. Our classification results on Hollywood2, KTH and UCF Sports are competitive with previously published results. To highlight some, on the KTH dataset, our recognition rate shows approximately 1% improvement in comparison to state-of-the-art methods even without supervision or dense sampling.

Part-based visual tracking is advantageous due to its ro-bustness against partial occlusion. However, how to effec-tively exploit the confidence scores of individual parts to construct a robust tracker is still a challenging problem. In this paper, we address this problem by simultaneously matching parts in each of multiple frames, which is realized by a locality-constrained low-rank sparse learning method that establishes multi-frame part correspondences through optimization of partial permutation matrices. The proposed part matching tracker (PMT) has a number of attractive properties. (1) It exploits the spatial-temporal locality-constrained property for robust part matching. (2) It match-es local parts from multiple frames jointly by considering their low-rank and sparse structure information, which can effectively handle part appearance variations due to occlu-sion or noise. (3) The proposed PMT model has the inbuilt mechanism of leveraging multi-mode target templates, so that the dilemma of template updating when encountering occlusion in tracking can be better handled. This contrasts with existing methods that only do part matching between a pair of frames. We evaluate PMT and compare with 10 pop-ular state-of-the-art methods on challenging benchmarks. Experimental results show that PMT consistently outperfor-m these existing trackers. 1.

We study in this paper the problem of learning classifiers from ambiguously labeled images. For instance, in the collection of new images, each image contains some samples of interest (\emph{e.g.,} human faces), and its associated caption has labels with the true ones included, while the sample-label association is unknown. The task is to learn classifiers from these ambiguously labeled images and generalize to new images. An essential consideration here is how to make use of the information embedded in the relations between samples and labels, both within each image and across the image set. To this end, we propose a novel framework to address this problem. Our framework is motivated by the observation that samples from the same class repetitively appear in the collection of ambiguously labeled training images, while they are just ambiguously labeled in each image. If we can identify samples of the same class from each image and associate them across the image set, the matrix formed by the samples from the same class would be ideally low-rank. By leveraging such a low-rank assumption, we can simultaneously optimize a partial permutation matrix (PPM) for each image, which is formulated in order to exploit all information between samples and labels in a principled way. The obtained PPMs can be readily used to assign labels to samples in training images, and then a standard SVM classifier can be trained and used for unseen data. Experiments on benchmark datasets show the effectiveness of our proposed method.

Promoting the spatial resolution of off-the-shelf hyperspectral sensors is expected to improve typical computer vision tasks, such as target tracking and image classification. In this paper, we investigate the scenario in which two cameras, one with a conventional RGB sensor and the other with a hyperspectral sensor, capture the same scene, attempting to extract redundant and complementary information. We propose a non-negative sparse promoting framework to integrate the hyperspectral and RGB data into a high resolution hyperspectral set of data. The formulated problem is in the form of a sparse non-negative matrix factorization with prior knowledge on the spectral and spatial transform responses, and it can be handled by alternating optimization where each subproblem is solved by efficient convex optimization solvers; e.g., the alternating direction method of multipliers. Experiments on a public database show that our method achieves much lower average reconstruction errors than other state-of-the-art methods.

Sharing knowledge for multiple related machine learning tasks is an effective strategy to improve the generalization performance. In this paper, we investigate knowledge sharing across categories for action recognition in videos. The motivation is that many action categories are related, where common motion pattern are shared among them (e.g. diving and high jump share the jump motion). We propose a new multi-task learning method to learn latent tasks shared across categories, and reconstruct a classifier for each category from these latent tasks. Compared to previous methods, our approach has two advantages: (1) The learned latent tasks correspond to basic motion patterns instead of full actions, thus enhancing discrimination power of the classifiers. (2) Categories are selected to share information with a sparsity regularizer, avoiding falsely forcing all categories to share knowledge. Experimental results on multiple public data sets show that the proposed approach can effectively transfer knowledge between different action categories to improve the performance of conventional single task learning methods.

In this paper, we propose a framework of transforming images from a source image space to a target image space, based on learning coupled dictionaries from a training set of paired images. The framework can be used for applications such as image super-resolution and estimation of image intrinsic components (shading and albedo). It is based on a local parametric regression approach, using sparse feature representations over learned coupled dictionaries across the source and target image spaces. After coupled dictionary learning, sparse coefficient vectors of training image patch pairs are partitioned into easily retrievable local clusters. For any test image patch, we can fast index into its closest local cluster and perform a local parametric regression between the learned sparse feature spaces. The obtained sparse representation (together with the learned target space dictionary) provides multiple constraints for each pixel of the target image to be estimated. The final target image is reconstructed based on these constraints. The contributions of our proposed framework are three-fold. 1) We propose a concept of coupled dictionary learning based on coupled sparse coding which requires the sparse coefficient vectors of a pair of corresponding source and target image patches to have the same support, i.e., the same indices of nonzero elements. 2) We devise a space partitioning scheme to divide the high-dimensional but sparse feature space into local clusters. The partitioning facilitates extremely fast retrieval of closest local clusters for query patches. 3) Benefiting from sparse feature-based image transformation, our method is more robust to corrupted input data, and can be considered as a simultaneous image restoration and transformation process. Experiments on intrinsic image estimation and super-resolution demonstrate the effectiveness and efficiency of our proposed method.

Optical flow estimation is a fundamental and ill-posed problem in computer vision. To recover a dense flow field, appropriate spatial constraints have to be enforced. Recent advances exploit higher order spatial regularization, and achieve the top performance on the Middlebury benchmark. In this work, we revisit learning-based approach, and propose a learned sparse model to patch-wisely regularize the flow field. In particular, our method is based on multi-scale spatial regularization, which benefits from first-order spatial regularity and our learned, higher order sparse model. To obtain accurate flow estimation, we propose a sequential optimization scheme to solve the corresponding energy minimization problem. Moreover, as the errors in intermediate flow estimates are usually dense with large variations, we further propose flow-driven and image-driven approaches to address the problem of outliers. Experiments on the Middlebury benchmark show that our method is competitive with the state-of-the-art.

Giving different human motions, there may be similar motion features in some body components, e.g., the motion features of arms when performing jogging and running, which are thus less discriminative for motion classification. In this paper, we consider counting less on these body components that have less discriminative information amongst different human motions. To this end, we present a new topic model, probabilistic latent semantic analysis based on multiple bags (M-PLSA), in which not all body components are considered equally important, i.e., motion features of less discriminative components are made less use of so that their contributions for classification are reduced. We use sparse spatio-temporal features extracted from videos to create visual words which are later assigned to different body components that they are detected from, so that co-occurrence matrices of different components can be calculated based on their corresponding vocabularies. Such label task can be automatically fulfilled by using the query visual words, i.e., words whose component labels are unknown, to traverse an automatic label tree (ALT) that grows from the training words with component labels. We show the performance of our approach on KTH dataset.

Gender is an important cue in social activities. In this correspondence, we present a study and analysis of gender classification based on human gait. Psychological experiments were carried out. These experiments showed that humans can recognize gender based on gait information, and that contributions of different body components vary. The prior knowledge extracted from the psychological experiments can be combined with an automatic method to further improve classification accuracy. The proposed method which combines human knowledge achieves higher performance than some other methods, and is even more accurate than human observers. We also present a numerical analysis of the contributions of different human components, which shows that head and hair, back, chest and thigh are more discriminative than other components. We also did challenging cross-race experiments that used Asian gait data to classify the gender of Europeans, and vice versa. Encouraging results were obtained. All the above prove that gait-based gender classification is feasible in controlled environments. In real applications, it still suffers from many difficulties, such as view variation, clothing and shoes changes, or carrying objects. We analyze the difficulties and suggest some possible solutions.

Human action video sequences can be considered as nonlinear dynamic shape manifolds in the space of image frames. In this paper, we address learning and classifying human actions on embedded low-dimensional manifolds. We propose a novel manifold embedding method, called Local Spatio-Temporal Discriminant Embedding (LSTDE). The discriminating capabilities of the proposed method are two-fold: (1) for local spatial discrimination, LSTDE projects data points (silhouette-based image frames of human action sequences) in a local neighborhood into the embedding space where data points of the same action class are close while those of different classes are far apart; (2) in such a local neighborhood, each data point has an associated short video segment, which forms a local temporal subspace on the embedded manifold. LSTDE finds an optimal embedding which maximizes the principal angles between those temporal subspaces associated with data points of different classes. Benefiting from the joint spatio-temporal discriminant embedding, our method is potentially more powerful for classifying human actions with similar space-time shapes, and is able to perform recognition on a frame-by-frame or short video segment basis. Experimental results demonstrate that our method can accurately recognize human actions, and can improve the recognition performance over some representative manifold embedding methods, especially on highly confusing human action types.

Existing learning-based face super-resolution (hallucination) techniques generate high-resolution images of a single facial modality (i.e., at a fixed expression, pose and illumination) given one or set of low-resolution face images as probe. Here, we present a generalized approach based on a hierarchical tensor (multilinear) space representation for hallucinating high-resolution face images across multiple modalities, achieving generalization to variations in expression and pose. In particular, we formulate a unified tensor which can be reduced to two parts: a global image-based tensor for modeling the mappings among different facial modalities, and a local patch-based multiresolution tensor for incorporating high-resolution image details. For realistic hallucination of unregistered low-resolution faces contained in raw images, we develop an automatic face alignment algorithm capable of pixel-wise alignment by iteratively warping the probing face to its projection in the space of training face images. Our experiments show not only performance superiority over existing benchmark face super-resolution techniques on single modal face hallucination, but also novelty of our approach in coping with multimodal hallucination and its robustness in automatic alignment under practical imaging conditions.

Existing approaches to learning-based face image super-resolution require low-resolution testing inputs manually registered to pre-aligned highresolution training models [9, 12, 13, 5]. This restricts automatic applications to live images and video. In this paper, we propose a multi-resolution patch tensor based model to automatically super-resolve and register low-resolution testing face images. Face candidates are triggered first by a face detector giving the subwindows with their coarse initial positions and scales in a large image frame. This initialises a combined registration and super-resolution process. Rather than manually aligning each coarsely detected face subwindow to some predefined template, based on its position and scale, we scan all the potential face subwindows across different positions and scales, and obtain registration and super-resolution in a simultaneous process. The superresolution result which is optimally correlated to its original low-resolution face subwindow is also guaranteed to be the best super-resolved reconstruction. We verify our approach by experimenting on MIT+CMU face detection dataset, the promising results demonstrate the robustness of our approach on learning-based face super-resolution on real images. 1

In this paper, we propose a sequential approach to hallucinate/ synthesize high-resolution images of multiple facial expressions. We propose an idea of multi-resolution tensor for super-resolution, and decompose facial expression images into small local patches. We build a multi-resolution patch tensor across different facial expressions. By unifying the identity parameters and learning the subspace mappings across different resolutions and expressions, we simplify the facial expression hallucination as a problem of parameter recovery in a patch tensor space. We further add a high-frequency component residue using nonparametric patch learning from high-resolution training data. We integrate the sequential statistical modelling into a Bayesian framework, so that given any low-resolution facial image of a single expression, we are able to synthesize multiple facial expression images in high-resolution. We show promising experimental results from both facial expression database and live video sequences.

Face images of non-frontal views under poor illumination with low resolution reduce dramatically face recognition accuracy. To overcome these problems, super-resolution techniques can be exploited. In this paper, we present a Bayesian framework to perform multi-modal (such as variations in viewpoint and illumination) face image super-resolutions in tensor space. Given a single modal low-resolution face image, we benefit from the multiple factor interactions of training tensor, and super-resolve its high-resolution reconstructions across different modalities. Instead of performing pixel-domain super-resolutions, we reconstruct the high-resolution face images by computing a maximum likelihood identity parameter vector in high-resolution tensor space. Experiments show promising results of multi-view and multi-illumination face image super-resolutions respectively.

In this paper, we present a novel learning-based algorithm to super-resolve multiple partially occluded low-resolution face images. By integrating hierarchical patch-wise alignment and inter-frame constraints into a Bayesian framework, we can probabilistically align multiple input images at different resolutions and recursively infer the high-resolution face image. We address the problem of fusing partial imagery information through multiple frames and discuss the new algorithm's effectiveness when encountering occluded low-re solution face images. We show promising results compared to that of existing face hallucination methods.

Face images of non-frontal views under poor illumination with low resolution reduce dramatically face recognition accuracy. To overcome these problems, super-resolution techniques can be exploited. In this paper, we present a Bayesian framework to perform multi-modal (such as variations in viewpoint and illumination) face image super-resolutions in tensor space. Given a single modal low-resolution face image, we benefit from the multiple factor interactions of training tensor, and super-resolve its high-resolution reconstructions across different modalities. Instead of performing pixel-domain super-resolutions, we reconstruct the high-resolution face images by computing a maximum likelihood identity parameter vector in high-resolution tensor space. Experiments show promising results of multi-view and multiillumination face image super-resolutions respectively. 1

Face images of non-frontal views under poor illumination resolution reduce dramatically face recognition accuracy. This is evident most compellingly by the very low recognition rate of all existing face recognition systems when applied to live CCTV camera input. In this paper, we present a Bayesian framework to perform multimodal (such as variations in viewpoint and illumination) face image super-resolution for recognition in tensor space. Given a single modal low-resolution face image, we benefit from the multiple factor interactions of training sensor and super-resolve its high-resolution reconstructions across different modalities for face recognition. Instead of performing pixel-domain super-resolution and recognition independently as two separate sequential processes, we integrate the tasks of super-resolution and recognition by directly computing a maximum likelihood identity parameter vector in high-resolution tensor space for recognition. We show results from multi-modal super-resolution and face recognition experiments across different imaging modalities, using low-resolution images as testing inputs and demonstrate improved recognition rates over standard tensorface and eigenface representations.