We introduce a novel representation for efficient classical rendering of photorealistic 3D face avatars. Leveraging recent advances in radiance fields anchored to parametric face models, our approach achieves controllable volumetric rendering of complex facial features, including hair, skin, and eyes. At enrollment time, we learn a set of radiance manifolds in 3D space to extract an explicit layered mesh, along with appearance and warp textures. During deployment, this allows us to control and animate the face through simple linear blending and alpha compositing of textures over a static mesh. This explicit representation also enables the generated avatar to be efficiently streamed online and then rendered using classical mesh and shader-based rendering on legacy graphics platforms, eliminating the need for any custom engineering or integration. https://syntec-research.github.io/LegacyAvatars/

Mixed Reality (MR) headsets promise a future of immersive telepresence where virtual humans blend indistinguishably into their real or virtual surroundings. Achieving this vision requires a method capable of capturing a user's motion, estimating their appearance under novel lighting, and understanding the surrounding environment - all from the constrained viewpoint of a head-mounted display (HMD). Existing approaches treat these as isolated problems: they either focus on driving avatars with baked-in lighting or rely on complex studio setups for relighting. In this paper, we present EgoRelight , a holistic framework for egocentric telepresence that simultaneously captures full-body human performance, synthesizes photorealistic and relightable appearance, and estimates high dynamic range (HDR) environment maps from a single HMD. First, to ensure accurate motion and surface reconstruction, we propose an egocentric perception module that leverages stereo down-facing cameras to extract dense depth maps, which serve as robust geometric control signals to drive a mesh-based avatar. Second, we introduce a novel neural appearance model that learns to synthesize view-dependent specular and view-independent diffuse shading separately. By employing a specialized ray-sampling strategy, our model generalizes to unseen illumination without relying on restrictive analytical BRDF priors. Third, we enable seamless avatar integration into the physical world via a test-time inverse rendering process, which recovers an HDR environment map by matching the pre-trained avatar's appearance to the live egocentric camera observations. We demonstrate the efficiency of our system through a social telepresence application, where remote users are coherently relit according to their local physical environment. Extensive experiments show that our individual components and the integrated system significantly outperform the best combinations of state-of-the-art baselines in both geometric accuracy and rendering as well as relighting fidelity. Further details can be found on our project page 1 .

Understanding of bimanual hand-object interaction plays an important role in robotics and virtual reality. However, due to significant occlusions between hands and object as well as the high degree-of-freedom motions, it is challenging to collect and annotate a high-quality, large-scale dataset, which prevents further improvement of bimanual hand-object interaction-related baselines. In this work, we propose a new 3D Gaussian Splatting based data augmentation framework for bimanual hand-object interaction, which is capable of augmenting existing dataset to large-scale photorealistic data with various hand-object pose and viewpoints. First, we use mesh-based 3DGS to model objects and hands, and to deal with the rendering blur problem due to multi-resolution input images used, we design a super-resolution module. Second, we extend the single hand grasping pose optimization module for the bimanual hand object to generate various poses of bimanual hand-object interaction, which can significantly expand the pose distribution of the dataset. Third, we conduct an analysis for the impact of different aspects of the proposed data augmentation on the understanding of the bimanual hand-object interaction. We perform our data augmentation on two benchmarks, H2O and Arctic, and verify that our method can improve the performance of the baselines.

Generating high-quality whole-body human object interaction motion sequences is becoming increasingly important in various fields such as animation, VR/AR, and robotics. The main challenge of this task lies in determining the level of involvement of each hand given the complex shapes of objects in different sizes and their different motion trajectories, while ensuring strong grasping realism and guaranteeing the coordination of movement in all body parts. Contrasting with existing work, which either generates human interaction motion sequences without detailed hand grasping poses or only models a static grasping pose, we propose a simple yet effective framework that jointly models the relationship between the body, hands, and the given object motion sequences within a single diffusion model. To guide our network in perceiving the object's spatial position and learning more natural grasping poses, we introduce novel contact-aware losses and incorporate a data-driven, carefully designed guidance. Experimental results demonstrate that our approach outperforms the state-of-the-art method and generates plausible results.

Visual programming has the potential of providing novice programmers with a low-code experience to build customized processing pipelines.Existing systems typically require users to build pipelines from scratch, implying that novice users are expected to set up and link appropriate nodes from a blank workspace.In this paper, we introduce InstructPipe, an AI assistant for prototyping machine learning (ML) pipelines with text instructions.We contribute two large language model (LLM) modules and a code interpreter as part of our framework.The LLM modules generate pseudocode for a target pipeline, and the interpreter renders the pipeline in the node-graph editor for further human-AI collaboration.Both technical and user evaluation (N=16) shows that InstructPipe empowers users to streamline their ML pipeline workfow, reduce their learning curve, and leverage open-ended commands to spark innovative ideas.

We propose a novel 3D-aware diffusion-based method for generating photorealistic talking head videos directly from a single identity image and explicit control signals (e.g., expressions). Our method generates Multiplane Images (MPIs) that ensure geometric consistency, making them ideal for immersive viewing experiences like binocular videos for VR headsets. Unlike existing methods that often require a separate stage or joint optimization to reconstruct a 3D representation (such as NeRF or 3D Gaussians), our approach directly generates the final output through a single denoising process, eliminating the need for post-processing steps to render novel views efficiently. To effectively learn from monocular videos, we introduce a training mechanism that reconstructs the output MPI randomly in either the target or the reference camera space. This approach enables the model to simultaneously learn sharp image details and underlying 3D information. Extensive experiments demonstrate the effectiveness of our method, which achieves competitive avatar quality and novel-view rendering capabilities, even without explicit 3D reconstruction or high-quality multi-view training data.

We present Exact Volumetric Ellipsoid Rendering (EVER), a method for real-time differentiable emission-only volume rendering. Unlike recent rasterization based approach by 3D Gaussian Splatting (3DGS), our primitive based representation allows for exact volume rendering, rather than alpha compositing 3D Gaussian billboards. As such, unlike 3DGS our formulation does not suffer from popping artifacts and view dependent density, but still achieves frame rates of $\sim\!30$ FPS at 720p on an NVIDIA RTX4090. Since our approach is built upon ray tracing it enables effects such as defocus blur and camera distortion (e.g. such as from fisheye cameras), which are difficult to achieve by rasterization. We show that our method is more accurate with fewer blending issues than 3DGS and follow-up work on view-consistent rendering, especially on the challenging large-scale scenes from the Zip-NeRF dataset where it achieves sharpest results among real-time techniques.

Video generation models have demonstrated great capability of producing impressive monocular videos, however, the generation of 3D stereoscopic video remains under-explored. We propose a pose-free and training-free approach for generating 3D stereoscopic videos using an off-the-shelf monocular video generation model. Our method warps a generated monocular video into camera views on stereoscopic baseline using estimated video depth, and employs a novel frame matrix video inpainting framework. The framework leverages the video generation model to inpaint frames observed from different timestamps and views. This effective approach generates consistent and semantically coherent stereoscopic videos without scene optimization or model fine-tuning. Moreover, we develop a disocclusion boundary re-injection scheme that further improves the quality of video inpainting by alleviating the negative effects propagated from disoccluded areas in the latent space. We validate the efficacy of our proposed method by conducting experiments on videos from various generative models, including Sora [4], Lumiere [2], WALT [8], and Zeroscope [12]. The experiments demonstrate that our method has a significant improvement over previous methods. Project page at https://daipengwa.github.io/SVG_ProjectPage/

Recently neural implicit rendering techniques have evolved rapidly and demonstrated significant advantages in novel view synthesis and 3D scene reconstruction. However, existing neural rendering methods for editing purposes offer limited functionalities, e.g., rigid transformation and category-specific editing. In this paper, we present a novel mesh-based representation by encoding the neural radiance field with disentangled geometry, texture, and semantic codes on mesh vertices, which empowers a set of efficient and comprehensive editing functionalities, including mesh-guided geometry editing, designated texture editing with texture swapping, filling and painting operations, and semantic-guided editing. To this end, we develop several techniques including a novel local space parameterization to enhance rendering quality and training stability, a learnable modification color on vertex to improve the fidelity of texture editing, a spatial-aware optimization strategy to realize precise texture editing, and a semantic-aided region selection to ease the laborious annotation of implicit field editing. Extensive experiments and editing examples on both real and synthetic datasets demonstrate the superiority of our method on representation quality and editing ability.

Virtual environments (VEs) are pivotal for virtual, augmented, and mixed reality systems. Despite advances in 3D generation and reconstruction, the direct creation of 3D objects within an established 3D scene (represented as NeRF) for novel VE creation remains a relatively unexplored domain. This process is complex, requiring not only the generation of high-quality 3D objects but also their seamless integration into the existing scene. To this end, we propose a novel pipeline featuring an intuitive interface, dubbed GO-NeRF. Our approach takes text prompts and user-specified regions as inputs and leverages the scene context to generate 3D objects within the scene. We employ a compositional rendering formulation that effectively integrates the generated 3D objects into the scene, utilizing optimized 3D-aware opacity maps to avoid unintended modifications to the original scene. Furthermore, we develop tailored optimization objectives and training strategies to enhance the model's ability to capture scene context and mitigate artifacts, such as floaters, that may occur while optimizing 3D objects within the scene. Extensive experiments conducted on both forward-facing and 360°scenes demonstrate the superior performance of our proposed method in generating objects that harmonize with surrounding scenes and synthesizing high-quality novel view images. The code will be at https://daipengwa.github.io/G0-NeRF/.

Remote video conferencing systems (RVCS) are widely adopted in personal and professional communication. However, they often lack the co-presence experience of in-person meetings. This is largely due to the absence of intuitive visual cues and clear spatial relationships among remote participants, which can lead to speech interruptions and loss of attention. This paper presents ChatDirector, a novel RVCS that overcomes these limitations by incorporating space-aware visual presence and speech-aware attention transition assistance. ChatDirector employs a real-time pipeline that converts participants’ RGB video streams into 3D portrait avatars and renders them in a virtual 3D scene. We also contribute a decision tree algorithm that directs the avatar layouts and behaviors based on participants’ speech states. We report on results from a user study (N=16) where we evaluated ChatDirector. The satisfactory algorithm performance and complimentary subject user feedback imply that ChatDirector significantly enhances communication efficacy and user engagement.

3D head avatars built with neural implicit volumetric representations have achieved unprecedented levels of pho-torealism. However, the computational cost of these methods remains a significant barrier to their widespread adoption, particularly in real-time applications such as virtual reality and teleconferencing. While attempts have been made to develop fast neural rendering approaches for static scenes, these methods cannot be simply employed to support realistic facial expressions, such as in the case of a dynamic facial performance. To address these challenges, we propose a novel fast 3D neural implicit head avatar model that achieves real-time rendering while maintaining fine-grained controllability and high rendering quality. Our key idea lies in the introduction of local hash table blendshapes, which are learned and attached to the vertices of an underlying face parametric model. These per-vertex hash-tables are linearly merged with weights predicted via a CNN, re-sulting in expression dependent embeddings. Our novel representation enables efficient density and color predictions using a lightweight MLP, which is further accelerated by a hierarchical nearest neighbor search method. Extensive experiments show that our approach runs in real-time while achieving comparable rendering quality to state-of-the-arts and decent results on challenging expressions.

Recently, we have witnessed the explosive growth of various volumetric representations in modeling animatable head avatars. However, due to the diversity of frameworks, there is no practical method to support high-level applications like 3D head avatar editing across different representations. In this paper, we propose a generic avatar editing approach that can be universally applied to various 3DMM-driving volumetric head avatars. To achieve this goal, we design a novel expression-aware modification generative model, which enables lift 2D editing from a single image to a consistent 3D modification field. To ensure the effectiveness of the generative modification process, we develop several techniques, including an expression-dependent modification distillation scheme to draw knowledge from the large-scale head avatar model and 2D facial texture editing tools, implicit latent space guidance to enhance model convergence, and a segmentation-based loss reweight strategy for fine-grained texture inversion. Extensive experiments demonstrate that our method delivers high-quality and consistent results across multiple expression and viewpoints. Project page: https://zju3dv.github.io/geneavatar/.

Pixel2Mesh (P2M) is a classical approach for reconstructing 3D shapes from a single color image through coarse-to-fine mesh deformation. Although P2M is capable of generating plausible global shapes, its Graph Convolution Network (GCN) often produces overly smooth results, causing the loss of fine-grained geometry details. Moreover, P2M generates non-credible features for occluded regions and struggles with the domain gap from synthetic data to real-world images, which is a common challenge for single-view 3D reconstruction methods. To address these challenges, we propose a novel Transformer-boosted architecture, named T-Pixel2Mesh, inspired by the coarse-to-fine approach of P2M. Specifically, we use a global Transformer to control the holistic shape and a local Transformer to progressively refine the local geometry details with graph-based point upsampling. To enhance real-world reconstruction, we present the simple yet effective Linear Scale Search (LSS), which serves as prompt tuning during the input preprocessing. Our experiments on ShapeNet demonstrate state-of-the-art performance, while results on real-world data show the generalization capability.

Structure from Motion (SfM) is a fundamental computer vision problem which has not been well handled by deep learning. One of the promising solutions is to apply explicit structural constraint, e.g., 3D cost volume, into the neural network. Obtaining accurate camera poses from images alone can be challenging, especially with complicated environmental factors. Existing methods usually assume accurate camera poses from GT or other methods, which is unrealistic in practice and additional sensors are needed. In this work, we design a physical driven architecture, namely DeepSFM, inspired by traditional Bundle Adjustment, which consists of two cost volume based architectures to iteratively refine depth and pose. The explicit constraints on both depth and pose, when combined with the learning components, bring merit from both traditional BA and emerging deep learning technology. To speed up the learning and inference efficiency, we apply the Gated Recurrent Units (GRUs)-based depth and pose update modules with coarse to fine cost volumes on the iterative refinements. In addition, with the extended residual depth prediction module, our model can be adapted to dynamic scenes effectively. Extensive experiments on various datasets show that our model achieves state-of-the-art performance with superior robustness against challenging inputs.

In recent years, there has been a proliferation of multimedia applications that leverage machine learning (ML) for interactive experiences. Prototyping ML-based applications is, however, still challenging, given complex workflows that are not ideal for design and experimentation. To better understand these challenges, we conducted a formative study with seven ML practitioners to gather insights about common ML evaluation workflows.

We propose a method to learn a high-quality implicit 3D head avatar from a monocular RGB video captured in the wild. The learnt avatar is driven by a parametric face model to achieve user-controlled facial expressions and head poses. Our hybrid pipeline combines the geometry prior and dynamic tracking of a 3DMM with a neural radiance field to achieve fine-grained control and photorealism. To reduce over-smoothing and improve out-of-model expressions synthesis, we propose to predict local features anchored on the 3DMM geometry. These learnt features are driven by 3DMM deformation and interpolated in 3D space to yield the volumetric radiance at a designated query point. We further show that using a Convolutional Neural Network in the UV space is critical in incorporating spatial context and producing representative local features. Extensive experiments show that we are able to reconstruct high-quality avatars, with more accurate expression-dependent details, good generalization to out-of-training expressions, and quantitatively superior renderings compared to other state-of-the-art approaches.

Despite the great success in 2D editing using user-friendly tools, such as Photoshop, semantic strokes, or even text prompts, similar capabilities in 3D areas are still limited, either relying on 3D modeling skills or allowing editing within only a few categories. In this paper, we present a novel semantic-driven NeRF editing approach, which enables users to edit a neural radiance field with a single image, and faithfully delivers edited novel views with high fidelity and multi-view consistency. To achieve this goal, we propose a prior-guided editing field to encode fine-grained geometric and texture editing in 3D space, and develop a series of techniques to aid the editing process, including cyclic constraints with a proxy mesh to facilitate geometric supervision, a color compositing mechanism to stabilize semantic-driven texture editing, and a feature-cluster-based regularization to preserve the irrelevant content unchanged. Extensive experiments and editing examples on both real-world and synthetic data demonstrate that our method achieves photo-realistic 3D editing using only a single edited image, pushing the bound of semantic-driven editing in 3D real-world scenes.

Rendering novel view images is highly desirable for many applications. Despite recent progress, it remains challenging to render high-fidelity and view-consistent novel views of large-scale scenes from in-the-wild images with inevitable artifacts (e.g., motion blur). To this end, we develop a hybrid neural rendering model that makes image-based representation and neural 3D representation join forces to render high-quality, view-consistent images. Besides, images captured in the wild inevitably contain artifacts, such as motion blur, which deteriorates the quality of rendered images. Accordingly, we propose strategies to simulate blur effects on the rendered images to mitigate the negative influence of blurriness images and reduce their importance during training based on precomputed quality-aware weights. Extensive experiments on real and synthetic data demonstrate our model surpasses state-of-the-art point-based methods for novel view synthesis. The code is available at https://daipengwa.github.io/Hybrid-Rendering-ProjectPage/.

Most existing point cloud upsampling methods have roughly three steps: feature extraction, feature expansion and 3D coordinate prediction. However, they usually suffer from two critical issues: (1) fixed upsampling rate after one-time training, since the feature expansion unit is customized for each upsampling rate; (2) outliers or shrinkage artifact caused by the difficulty of precisely predicting 3D coordinates or residuals of upsampled points. To adress them, we propose a new framework for accurate point cloud upsampling that supports arbitrary upsampling rates. Our method first interpolates the low-res point cloud according to a given upsampling rate. And then refine the positions of the interpolated points with an iterative optimization process, guided by a trained model estimating the difference between the current point cloud and the high-res target. Extensive quantitative and qualitative results on benchmarks and downstream tasks demonstrate that our method achieves the state-of-the-art accuracy and efficiency.

Light-weight time-of-flight (ToF) depth sensors are compact and cost-efficient, and thus widely used on mobile devices for tasks such as autofocus and obstacle detection. However, due to the sparse and noisy depth measurements, these sensors have rarely been considered for dense geometry reconstruction. In this work, we present the first dense SLAM system with a monocular camera and a light-weight ToF sensor. Specifically, we propose a multi-modal implicit scene representation that supports rendering both the signals from the RGB camera and light-weight ToF sensor which drives the optimization by comparing with the raw sensor inputs. Moreover, in order to guarantee successful pose tracking and reconstruction, we exploit a predicted depth as an intermediate supervision and develop a coarse-to-fine optimization strategy for efficient learning of the implicit representation. At last, the temporal information is explicitly exploited to deal with the noisy signals from light-weight ToF sensors to improve the accuracy and robustness of the system. Experiments demonstrate that our system well exploits the signals of light-weight ToF sensors and achieves competitive results both on camera tracking and dense scene reconstruction. Project page: https://zju3dv.github.io/tof_slam/.

Auto-Regressive (AR) models have achieved impressive results in 2D image generation by modeling joint distributions in the grid space. While this approach has been extended to the 3D domain for powerful shape generation, it still has two limitations: expensive computations on volumetric grids and ambiguous auto-regressive order along grid dimensions. To overcome these limitations, we propose the Improved Auto-regressive Model (ImAM) for 3D shape generation, which applies discrete representation learning based on a latent vector instead of volumetric grids. Our approach not only reduces computational costs but also preserves essential geometric details by learning the joint distribution in a more tractable order. Moreover, thanks to the simplicity of our model architecture, we can naturally extend it from unconditional to conditional generation by concatenating various conditioning inputs, such as point clouds, categories, images, and texts. Extensive experiments demonstrate that ImAM can synthesize diverse and faithful shapes of multiple categories, achieving state-of-the-art performance.

Hand-object interaction understanding and the barely addressed novel view synthesis are highly desired in the immersive communication, whereas it is challenging due to the high deformation of hand and heavy occlusions between hand and object. In this paper, we propose a neural rendering and pose estimation system for hand-object interaction from sparse views, which can also enable 3D hand-object interaction editing. We share the inspiration from recent scene understanding work that shows a scene specific model built beforehand can significantly improve and unblock vision tasks especially when inputs are sparse, and extend it to the dynamic hand-object interaction scenario and propose to solve the problem in two stages. We first learn the shape and appearance prior knowledge of hands and objects separately with the neural representation at the offline stage. During the online stage, we design a rendering-based joint model fitting framework to understand the dynamic hand-object interaction with the pre-built hand and object models as well as interaction priors, which thereby overcomes penetration and separation issues between hand and object and also enables novel view synthesis. In order to get stable contact during the hand-object interaction process in a sequence, we propose a stable contact loss to make the contact region to be consistent. Experiments demonstrate that our method outperforms the state-of-the-art methods. Code and dataset are available in project web-page https://iscas3dv.github.io/HO-NeRF.

We propose a novel transformer-based framework that reconstructs two high fidelity hands from multi-view RGB images. Unlike existing hand pose estimation methods, where one typically trains a deep network to regress hand model parameters from single RGB image, we consider a more challenging problem setting where we directly regress the absolute root poses of two-hands with extended forearm at high resolution from egocentric view. As existing datasets are either infeasible for egocentric viewpoints or lack background variations, we create a large-scale synthetic dataset with diverse scenarios and collect a real dataset from multi-calibrated camera setup to verify our proposed multi-view image feature fusion strategy. To make the reconstruction physically plausible, we propose two strategies: (i) a coarse-to-fine spectral graph convolution decoder to smoothen the meshes during upsampling and (ii) an optimisation-based refinement stage at inference to prevent self-penetrations. Through extensive quantitative and qualitative evaluations, we show that our framework is able to produce realistic two-hand reconstructions and demonstrate the generalisation of synthetic-trained models to real data, as well as real-time AR/VR applications.

Learning 3D shape representation with dense correspondence for deformable objects is a fundamental problem in computer vision. Existing approaches often need additional annotations of specific semantic domain, e.g., skeleton poses for human bodies or animals, which require extra annotation effort and suffer from error accumulation, and they are limited to specific domain. In this paper, we propose a novel self-supervised approach to learn neural implicit shape representation for deformable objects, which can represent shapes with a template shape and dense correspondence in 3D. Our method does not require the priors of skeleton and skinning weight, and only requires a collection of shapes represented in signed distance fields. To handle the large deformation, we constrain the learned template shape in the same latent space with the training shapes, design a new formulation of local rigid constraint that enforces rigid transformation in local region and addresses local reflection issue, and present a new hierarchical rigid constraint to reduce the ambiguity due to the joint learning of template shape and correspondences. Extensive experiments show that our model can represent shapes with large deformations. We also show that our shape representation can support two typical applications, such as texture transfer and shape editing, with competitive performance. The code and models are available at https://iscas3dv.github.io/deformshape.

High-fidelity, photorealistic 3D capture of a human face is a long-standing problem in computer graphics – the complex material of skin, intricate geometry of hair, and fine scale textural details make it challenging. Traditional techniques rely on very large and expensive capture rigs to reconstruct explicit mesh geometry and appearance maps, and are limited by the accuracy of hand-crafted reflectance models. More recent volumetric methods (e.g., NeRFs) have enabled view-synthesis and sometimes relighting by learning an implicit representation of the density and reflectance basis, but suffer from artifacts and blurriness due to the inherent ambiguities in volumetric modeling. These problems are further exacerbated when capturing with few cameras and light sources. We present a novel technique for high-quality capture of a human face for 3D view synthesis and relighting using a sparse, compact capture rig consisting of 15 cameras and 15 lights. Our method combines a neural volumetric representation with traditional mesh reconstruction from multiview stereo. The proxy geometry allows us to anchor the 3D density field to prevent artifacts and guide the disentanglement of intrinsic radiance components of the face appearance such as diffuse and specular reflectance, and incident radiance (shadowing) fields. Our hybrid representation significantly improves the state-of-the-art quality for arbitrarily dense renders of a face from desired camera viewpoint as well as environmental, directional, and near-field lighting.

3D hand pose estimation is a challenging problem in computer vision due to the high degrees-of-freedom of hand articulated motion space and large viewpoint variation. As a consequence, similar poses observed from multiple views can be dramatically different. In order to deal with this issue, view-independent features are required to achieve state-of-the-art performance. In this paper, we investigate the impact of view-independent features on 3D hand pose estimation from a single depth image, and propose a novel recurrent neural network for 3D hand pose estimation, in which a cascaded 3D pose-guided alignment strategy is designed for view-independent feature extraction and a recurrent hand pose module is designed for modeling the dependencies among sequential aligned features for 3D hand pose estimation. In particular, our cascaded pose-guided 3D alignments are performed in 3D space in a coarse-to-fine fashion. First, hand joints are predicted and globally transformed into a canonical reference frame; Second, the palm of the hand is detected and aligned; Third, local transformations are applied to the fingers to refine the final predictions. The proposed recurrent hand pose module for aligned 3D representation can extract recurrent pose-aware features and iteratively refines the estimated hand pose. Our recurrent module could be utilized for both single-view estimation and sequence-based estimation with 3D hand pose tracking. Experiments show that our method improves the state-of-the-art by a large margin on popular benchmarks with the simple yet efficient alignment and network architectures.

Despite the impressive results achieved by deep learning based 3D reconstruction, the techniques of directly learning to model 4D human captures with detailed geometry have been less studied. This work presents a novel neural compositional representation for Human 4D Modeling with transformER (H4MER). Specifically, our H4MER is a compact and compositional representation for dynamic human by exploiting the human body prior from the widely used SMPL parametric model. Thus, H4MER can represent a dynamic 3D human over a temporal span with the codes of shape, initial pose, motion and auxiliaries. A simple yet effective linear motion model is proposed to provide a rough and regularized motion estimation, followed by per-frame compensation for pose and geometry details with the residual encoded in the auxiliary codes. We present a novel Transformer-based feature extractor and conditional GRU decoder to facilitate learning and improve the representation capability. Extensive experiments demonstrate our method is not only effective in recovering dynamic human with accurate motion and detailed geometry, but also amenable to various 4D human related tasks, including monocular video fitting, motion retargeting, 4D completion, and future prediction.

We study the problem of shape generation in 3D mesh representation from a small number of color images with or without camera poses. While many previous works learn to hallucinate the shape directly from priors, we adopt to further improve the shape quality by leveraging cross-view information with a graph convolution network. Instead of building a direct mapping function from images to 3D shape, our model learns to predict series of deformations to improve a coarse shape iteratively. Inspired by traditional multiple view geometry methods, our network samples nearby area around the initial mesh's vertex locations and reasons an optimal deformation using perceptual feature statistics built from multiple input images. Extensive experiments show that our model produces accurate 3D shapes that are not only visually plausible from the input perspectives, but also well aligned to arbitrary viewpoints. With the help of physically driven architecture, our model also exhibits generalization capability across different semantic categories, and the number of input images. Model analysis experiments show that our model is robust to the quality of the initial mesh and the error of camera pose, and can be combined with a differentiable renderer for test-time optimization.

3D hand pose estimation from single depth is a fundamental problem in computer vision, and has wide applications. However, the existing methods still can not achieve satisfactory hand pose estimation results due to view variation and occlusion of human hand. In this paper, we propose a new virtual view selection and fusion module for 3D hand pose estimation from single depth. We propose to automatically select multiple virtual viewpoints for pose estimation and fuse the results of all and find this empirically delivers accurate and robust pose estimation. In order to select most effective virtual views for pose fusion, we evaluate the virtual views based on the confidence of virtual views using a light-weight network via network distillation. Experiments on three main benchmark datasets including NYU, ICVL and Hands2019 demonstrate that our method outperforms the state-of-the-arts on NYU and ICVL, and achieves very competitive performance on Hands2019-Task1, and our proposed virtual view selection and fusion module is both effective for 3D hand pose estimation.

Local density of point clouds is crucial for representing local details, but has been overlooked by existing point cloud compression methods. To address this, we propose a novel deep point cloud compression method that preserves local density information. Our method works in an auto-encoder fashion: the encoder downsamples the points and learns point-wise features, while the decoder upsamples the points using these features. Specifically, we propose to encode local geometry and density with three embeddings: density embedding, local position embedding and ancestor embedding. During the decoding, we explicitly predict the upsampling factor for each point, and the directions and scales of the upsampled points. To mitigate the clustered points issue in existing methods, we design a novel sub-point convolution layer, and an upsampling block with adaptive scale. Furthermore, our method can also compress point-wise attributes, such as normal. Extensive qualitative and quantitative results on SemanticKITTI and ShapeNet demonstrate that our method achieves the state-of-the-art rate-distortion trade-off.

Despite the impressive results achieved by deep learning based 3D reconstruction, the techniques of directly learning to model 4D human captures with detailed geometry have been less studied. This work presents a novel framework that can effectively learn a compact and compositional representation for dynamic human by exploiting the human body prior from the widely used SMPL parametric model. Particularly, our representation, named H4D, represents a dynamic 3D human over a temporal span with the SMPL parameters of shape and initial pose, and latent codes encoding motion and auxiliary information. A simple yet effective linear motion model is proposed to provide a rough and regularized motion estimation, followed by perframe compensation for pose and geometry details with the residual encoded in the auxiliary code. Technically, we introduce novel GRU-based architectures to facilitate learning and improve the representation capability. Extensive experiments demonstrate our method is not only efficacy in recovering dynamic human with accurate motion and detailed geometry, but also amenable to various 4D human related tasks, including motion retargeting, motion completion and future prediction.

We, as human beings, can understand and picture a familiar scene from arbitrary viewpoints given a single image, whereas this is still a grand challenge for computers. We hereby present a novel solution to mimic such human perception capability based on a new paradigm of amodal 3D scene understanding with neural rendering for a closed scene. Specifically, we first learn the prior knowledge of the objects in a closed scene via an offline stage, which facilitates an online stage to understand the room with unseen furniture arrangement. During the online stage, given a panoramic image of the scene in different layouts, we utilize a holistic neural-rendering-based optimization framework to efficiently estimate the correct 3D scene layout and deliver realistic free-viewpoint rendering. In order to handle the domain gap between the offline and online stage, our method exploits compositional neural rendering techniques for data augmentation in the offline training. The experiments on both synthetic and real datasets demonstrate that our two-stage design achieves robust 3D scene understanding and outperforms competing methods by a large margin, and we also show that our realistic free-viewpoint rendering enables various applications, including scene touring and editing. Code and data are available on the project webpage: https://zju3dv.github.io/nr_in_a_room/.

Learning based representation has become the key to the success of many computer vision systems. While many 3D representations have been proposed, it is still an unaddressed problem how to represent a dynamically changing 3D object. In this paper, we introduce a compositional representation for 4D captures, i.e. a deforming 3D object over a temporal span, that disentangles shape, initial state, and motion respectively. Each component is represented by a latent code via a trained encoder. To model the motion, a neural Ordinary Differential Equation (ODE) is trained to update the initial state conditioned on the learned motion code, and a decoder takes the shape code and the updated state code to reconstruct the 3D model at each time stamp. To this end, we propose an Identity Exchange Training (IET) strategy to encourage the network to learn effectively decoupling each component. Extensive experiments demonstrate that the proposed method outperforms existing state-of-the-art deep learning based methods on 4D reconstruction, and significantly improves on various tasks, including motion transfer and completion.

In this paper, we address the problem of building dense correspondences between human images under arbitrary camera viewpoints and body poses. Prior art either assumes small motion between frames or relies on local descriptors, which cannot handle large motion or visually ambiguous body parts, e.g., left vs. right hand. In contrast, we propose a deep learning framework that maps each pixel to a feature space, where the feature distances reflect the geodesic distances among pixels as if they were projected onto the surface of a 3D human scan. To this end, we introduce novel loss functions to push features apart according to their geodesic distances on the surface. Without any semantic annotation, the proposed embeddings automatically learn to differentiate visually similar parts and align different subjects into an unified feature space. Extensive experiments show that the learned embeddings can produce accurate correspondences between images with remarkable generalization capabilities on both intra and inter subjects.1

This paper presents HITNet, a novel neural network architecture for real-time stereo matching. Contrary to many recent neural network approaches that operate on a full cost volume and rely on 3D convolutions, our approach does not explicitly build a volume and instead relies on a fast multi-resolution initialization step, differentiable 2D geometric propagation and warping mechanisms to infer disparity hypotheses. To achieve a high level of accuracy, our network not only geometrically reasons about disparities but also infers slanted plane hypotheses allowing to more accurately perform geometric warping and upsampling operations. Our architecture is inherently multi-resolution allowing the propagation of information across different levels. Multiple experiments prove the effectiveness of the proposed approach at a fraction of the computation required by state-of-the-art methods. At the time of writing, HITNet ranks 1st-3rdon all the metrics published on the ETH3D website for two view stereo, ranks 1ston most of the metrics amongst all the end-to-end learning approaches on Middlebury-v3, ranks 1ston the popular KITTI 2012 and 2015 benchmarks among the published methods faster than 100 ms.

We present a new pipeline for holistic 3D scene understanding from a single image, which could predict object shapes, object poses, and scene layout. As it is a highly ill-posed problem, existing methods usually suffer from inaccurate estimation of both shapes and layout especially for the cluttered scene due to the heavy occlusion between objects. We propose to utilize the latest deep implicit representation to solve this challenge. We not only propose an image-based local structured implicit network to improve the object shape estimation, but also refine the 3D object pose and scene layout via a novel implicit scene graph neural network that exploits the implicit local object features. A novel physical violation loss is also proposed to avoid incorrect context between objects. Extensive experiments demonstrate that our method outperforms the state-of-the-art methods in terms of object shape, scene layout estimation, and 3D object detection.

We present SOLID-Net, a neural network for spatially- varying outdoor lighting estimation from a single outdoor image for any 2D pixel location. Previous work has used a unified sky environment map to represent outdoor lighting. Instead, we generate spatially-varying local lighting environment maps by combining global sky environment map with warped image information according to geometric information estimated from intrinsics. As no outdoor dataset with image and local lighting ground truth is readily available, we introduce the SOLID-Img dataset with physically- based rendered images and their corresponding intrinsic and lighting information. We train a deep neural network to regress intrinsic cues with physically-based constraints and use them to conduct global and local lightings estimation. Experiments on both synthetic and real datasets show that SOLID-Net significantly outperforms previous methods.

We introduce Multiresolution Deep Implicit Functions (MDIF), a hierarchical representation that can recover fine geometry detail, while being able to perform global operations such as shape completion. Our model represents a complex 3D shape with a hierarchy of latent grids, which can be decoded into different levels of detail and also achieve better accuracy. For shape completion, we propose latent grid dropout to simulate partial data in the latent space and therefore defer the completing functionality to the decoder side. This along with our multires design significantly improves the shape completion quality under decoder-only latent optimization. To the best of our knowledge, MDIF is the first deep implicit function model that can at the same time (1) represent different levels of detail and allow progressive decoding; (2) support both encoder-decoder inference and decoder-only latent optimization, and fulfill multiple applications; (3) perform detailed decoder-only shape completion. Experiments demonstrate its superior performance against prior art in various 3D reconstruction tasks.

We present a deep learning pipeline that leverages network self-prior to recover a full 3D model consisting of both a triangular mesh and a texture map from the colored 3D point cloud. Different from previous methods either exploiting 2D self-prior for image editing or 3D self-prior for pure surface reconstruction, we propose to exploit a novel hybrid 2D-3D self-prior in deep neural networks to significantly improve the geometry quality and produce a high-resolution texture map, which is typically missing from the output of commodity-level 3D scanners. In particular, we first generate an initial mesh using a 3D convolutional neural network with 3D self-prior, and then encode both 3D information and color information in the 2D UV atlas, which is further refined by 2D convolutional neural networks with the self-prior. In this way, both 2D and 3D self-priors are utilized for the mesh and texture recovery. Experiments show that, without the need of any additional training data, our method recovers the 3D textured mesh model of high quality from sparse input, and outperforms the state-of-the-art methods in terms of both geometry and texture quality.

Implicit neural rendering techniques have shown promising results for novel view synthesis. However, existing methods usually encode the entire scene as a whole, which is generally not aware of the object identity and limits the ability to the high-level editing tasks such as moving or adding furniture. In this paper, we present a novel neural scene rendering system, which learns an object-compositional neural radiance field and produces realistic rendering with editing capability for a clustered and real-world scene. Specifically, we design a novel two-pathway architecture, in which the scene branch encodes the scene geometry and appearance, and the object branch encodes each standalone object conditioned on learnable object activation codes. To survive the training in heavily cluttered scenes, we propose a scene-guided training strategy to solve the 3D space ambiguity in the occluded regions and learn sharp boundaries for each object. Extensive experiments demonstrate that our system not only achieves competitive performance for static scene novel-view synthesis, but also produces realistic rendering for object-level editing.

Panorama images have a much larger field-of-view thus naturally encode enriched scene context information compared to standard perspective images, which however is not well exploited in the previous scene understanding methods. In this paper, we propose a novel method for panoramic 3D scene understanding which recovers the 3D room layout and the shape, pose, position, and semantic category for each object from a single full-view panorama image. In order to fully utilize the rich context information, we design a novel graph neural network based context model to predict the relationship among objects and room layout, and a differentiable relationship-based optimization module to optimize object arrangement with well-designed objective functions on-the-fly. Realizing the existing data are either with incomplete ground truth or overly-simplified scene, we present a new synthetic dataset with good diversity in room layout and furniture placement, and realistic image quality for total panoramic 3D scene understanding. Experiments demonstrate that our method outperforms existing methods on panoramic scene understanding in terms of both geometry accuracy and object arrangement. Code is available at https://chengzhag.github.io/publication/dpc.

In this paper, we propose a novel deep learning framework to reconstruct 3D hand poses and shapes of two interacting hands from a single color image. Previous methods designed for single hand cannot be easily applied for the two hand scenario because of the heavy inter-hand occlusion and larger solution space. In order to address the occlusion and similar appearance between hands that may confuse the network, we design a hand pose-aware attention module to extract features associated to each individual hand respectively. We then leverage the two hand context presented in interaction to propose a context-aware cascaded refinement that improves the hand pose and shape accuracy of each hand conditioned on the context between interacting hands. Extensive experiments on the main benchmark datasets demonstrate that our method predicts accurate 3D hand pose and shape from single color image, and achieves the state-of-the-art performance. Code is available in project webpage https://baowenz.github.io/Intershape/.

In this paper, we propose an end-to-end deep learning architecture that generates 3D triangular meshes from single color images. Restricted by the nature of prevalent deep learning techniques, the majority of previous works represent 3D shapes in volumes or point clouds. However, it is non-trivial to convert these representations to compact and ready-to-use mesh models. Unlike the existing methods, our network represents 3D shapes in meshes, which are essentially graphs and well suited for graph-based convolutional neural networks. Leveraging perceptual features extracted from an input image, our network produces the correct geometry by progressively deforming an ellipsoid. To make the whole deformation procedure stable, we adopt a coarse-to-fine strategy, and define various mesh/surface related losses to capture properties of various aspects, which benefits producing the visually appealing and physically accurate 3D geometry. In addition, our model by nature can be adapted to objects in specific domains, e.g., human faces, and be easily extended to learn per-vertex properties, e.g., color. Extensive experiments show that our method not only qualitatively produces the mesh model with better details, but also achieves the higher 3D shape estimation accuracy compared against the state-of-the-arts.

Hand pose understanding is essential to applications such as human computer interaction and augmented reality. Recently, deep learning based methods achieve great progress in this problem. However, the lack of high-quality and large-scale dataset prevents the further improvement of hand pose related tasks such as 2D/3D hand pose from color and depth from color. In this paper, we develop a large-scale and high-quality synthetic dataset, PBRHand. The dataset contains millions of photo-realistic rendered hand images and various ground truths including pose, semantic segmentation, and depth. Based on the dataset, we firstly investigate the effect of rendering methods and used databases on the performance of three hand pose related tasks: 2D/3D hand pose from color, depth from color and 3D hand pose from depth. This study provides insights that photo-realistic rendering dataset is worthy of synthesizing and shows that our new dataset can improve the performance of the state-of-the-art on these tasks. This synthetic data also enables us to explore multi-task learning, while it is expensive to have all the ground truth available on real data. Evaluations show that our approach can achieve state-of-the-art or competitive performance on several public datasets.

Recent emerging technologies such AR/VR and HCI are drawing high demand on more comprehensive hand shape understanding, requiring not only 3D hand skeleton pose but also hand shape geometry. In this paper, we propose a deep learning framework to produce 3D hand shape from a single depth image. To address the challenge that capturing ground truth 3D hand shape in the training dataset is non-trivial, we leverage synthetic data to construct a statistical hand shape model and adopt weak supervision from widely accessible hand skeleton pose annotation. To bridge the gap due to the different hand skeleton definitions in the existing public datasets, we propose a joint regression network for hand pose adaptation. To reconstruct the hand shape, we use Chamfer loss between the predicted hand shape and the point cloud from the input depth to learn the shape reconstruction model in a weakly-supervised manner. Experiments demonstrate that our model adapts well to the real data and produces accurate hand shapes that outperform the state-of-the-art methods both qualitatively and quantitatively.

We present a new deep point cloud rendering pipeline through multi-plane projections. The input to the network is the raw point cloud of a scene and the output are image or image sequences from a novel view or along a novel camera trajectory. Unlike previous approaches that directly project features from 3D points onto 2D image domain, we propose to project these features into a layered volume of camera frustum. In this way, the visibility of 3D points can be automatically learnt by the network, such that ghosting effects due to false visibility check as well as occlusions caused by noise interferences are both avoided successfully. Next, the 3D feature volume is fed into a 3D CNN to produce multiple planes of images w.r.t. the space division in the depth directions. The multi-plane images are then blended based on learned weights to produce the final rendering results. Experiments show that our network produces more stable renderings compared to previous methods, especially near the object boundaries. Moreover, our pipeline is robust to noisy and relatively sparse point cloud for a variety of challenging scenes.

We propose a differentiable sphere tracing algorithm to bridge the gap between inverse graphics methods and the recently proposed deep learning based implicit signed distance function. Due to the nature of the implicit function, the rendering process requires tremendous function queries, which is particularly problematic when the function is represented as a neural network. We optimize both the forward and backward pass of our rendering layer to make it run efficiently with affordable memory consumption on a commodity graphics card. Our rendering method is fully differentiable such that losses can be directly computed on the rendered 2D observations, and the gradients can be propagated backward to optimize the 3D geometry. We show that our rendering method can effectively reconstruct accurate 3D shapes from various inputs, such as sparse depth and multi-view images, through inverse optimization. With the geometry based reasoning, our 3D shape prediction methods show excellent generalization capability and robustness against various noises.

We describe a novel approach for compressing truncated signed distance fields (TSDF) stored in 3D voxel grids, and their corresponding textures. To compress the TSDF, our method relies on a block-based neural network architecture trained end-to-end, achieving state-of-the-art rate-distortion trade-off. To prevent topological errors, we losslessly com- press the signs of the TSDF, which also upper bounds the reconstruction error by the voxel size. To compress the corresponding texture, we designed a fast block-based UV parameterization, generating coherent texture maps that can be effectively compressed using existing video compression algorithms. We demonstrate the performance of our algo- rithms on two 4D performance capture datasets, reducing bitrate by 66% for the same distortion, or alternatively re- ducing the distortion by 50% for the same bitrate, compared to the state-of-the-art.

Pose transfer has been studied for decades, in which the pose of a source mesh is applied to a target mesh. Particularly in this paper, we are interested in transferring the pose of source human mesh to deform the target human mesh, while the source and target meshes may have different identity information. Traditional studies assume that the paired source and target meshes are existed with the point-wise correspondences of user annotated landmarks/mesh points, which requires heavy labelling efforts. On the other hand, the generalization ability of deep models is limited, when the source and target meshes have different identities. To break this limitation, we proposes the first neural pose transfer model that solves the pose transfer via the latest technique for image style transfer, leveraging the newly proposed component -- spatially adaptive instance normalization. Our model does not require any correspondences between the source and target meshes. Extensive experiments show that the proposed model can effectively transfer deformation from source to target meshes, and has good generalization ability to deal with unseen identities or poses of meshes. Code is available at https://github.com/jiashunwang/Neural-Pose-Transfer.

Physically based rendering has been widely used to generate photo-realistic images, which greatly impacts industry by providing appealing rendering, such as for entertainment and augmented reality, and academia by serving large scale high-fidelity synthetic training data for data hungry methods like deep learning. However, physically based rendering heavily relies on ray-tracing, which can be computational expensive in complicated environment and hard to parallelize. In this paper, we propose an end-to-end deep learning based approach to generate physically based rendering efficiently. Our system consists of two stacked neural networks, which effectively simulates the physical behavior of the rendering process and produces photo-realistic images. The first network, namely shading network, is designed to predict the optimal shading image from surface normal, depth and illumination; the second network, namely composition network, learns to combine the predicted shading image with the reflectance to generate the final result. Our approach is inspired by intrinsic image decomposition, and thus it is more physically reasonable to have shading as intermediate supervision. Extensive experiments show that our approach is robust to noise thanks to a modified perceptual loss and even outperforms the physically based rendering systems in complex scenes given a reasonable time budget.

The increasing demand for 3D content in augmented and virtual reality has motivated the development of volumetric performance capture systemsnsuch as the Light Stage. Recent advances are pushing free viewpoint relightable videos of dynamic human performances closer to photorealistic quality. However, despite significant efforts, these sophisticated systems are limited by reconstruction and rendering algorithms which do not fully model complex 3D structures and higher order light transport effects such as global illumination and sub-surface scattering. In this paper, we propose a system that combines traditional geometric pipelines with a neural rendering scheme to generate photorealistic renderings of dynamic performances under desired viewpoint and lighting. Our system leverages deep neural networks that model the classical rendering process to learn implicit features that represent the view-dependent appearance of the subject independent of the geometry layout, allowing for generalization to unseen subject poses and even novel subject identity. Detailed experiments and comparisons demonstrate the efficacy and versatility of our method to generate high-quality results, significantly outperforming the existing state-of-the-art solutions.

In this paper, we propose a deep learning architecture that produces accurate dense depth for the outdoor scene from a single color image and a sparse depth. Inspired by the indoor depth completion, our network estimates surface normals as the intermediate representation to produce dense depth, and can be trained end-to-end. With a modified encoder-decoder structure, our network effectively fuses the dense color image and the sparse LiDAR depth. To address outdoor specific challenges, our network predicts a confidence mask to handle mixed LiDAR signals near foreground boundaries due to occlusion, and combines estimates from the color image and surface normals with learned attention maps to improve the depth accuracy especially for distant areas. Extensive experiments demonstrate that our model improves upon the state-of-the-art performance on KITTI depth completion benchmark. Ablation study shows the positive impact of each model components to the final performance, and comprehensive analysis shows that our model generalizes well to the input with higher sparsity or from indoor scenes.

We study the problem of shape generation in 3D mesh representation from a few color images with known camera poses. While many previous works learn to hallucinate the shape directly from priors, we resort to further improving the shape quality by leveraging cross-view information with a graph convolutional network. Instead of building a direct mapping function from images to 3D shape, our model learns to predict series of deformations to improve a coarse shape iteratively. Inspired by traditional multiple view geometry methods, our network samples nearby area around the initial mesh's vertex locations and reasons an optimal deformation using perceptual feature statistics built from multiple input images. Extensive experiments show that our model produces accurate 3D shape that are not only visually plausible from the input perspectives, but also well aligned to arbitrary viewpoints. With the help of physically driven architecture, our model also exhibits generalization capability across different semantic categories, number of input images, and quality of mesh initialization.

The ability to quickly recognize and learn new visual concepts from limited samples enables humans to quickly adapt to new tasks and environments. This ability is enabled by semantic association of novel concepts with those that have already been learned and stored in memory. Computers can start to ascertain similar abilities by utilizing a semantic concept space. A concept space is a high-dimensional semantic space in which similar abstract concepts appear close and dissimilar ones far apart. In this paper, we propose a novel approach to one-shot learning that builds on this core idea. Our approach learns to map a novel sample instance to a concept, relates that concept to the existing ones in the concept space and, using these relationships, generates new instances, by interpolating among the concepts, to help learning. Instead of synthesizing new image instance, we propose to directly synthesize instance features by leveraging semantics using a novel auto-encoder network we call dual TriNet. The encoder part of the TriNet learns to map multi-layer visual features from CNN to a semantic vector. In semantic space, we search for related concepts, which are then projected back into the image feature spaces by the decoder portion of the TriNet. Two strategies in the semantic space are explored. Notably, this seemingly simple strategy results in complex augmented feature distributions in the image feature space, leading to substantially better performance. The codes and models are released in the github: https://github.com/tankche1/ Semantic-Feature-Augmentation-in-Few-shot-Learning.

The goal of our work is to complete the depth channel of an RGB-D image. Commodity-grade depth cameras often fail to sense depth for shiny, bright, transparent, and distant surfaces. To address this problem, we train a deep network that takes an RGB image as input and predicts dense surface normals and occlusion boundaries. Those predictions are then combined with raw depth observations provided by the RGB-D camera to solve for depths for all pixels, including those missing in the original observation. This method was chosen over others (e.g., inpainting depths directly) as the result of extensive experiments with a new depth completion benchmark dataset, where holes are filled in training data through the rendering of surface reconstructions created from multiview RGB-D scans. Experiments with different network inputs, depth representations, loss functions, optimization methods, inpainting methods, and deep depth estimation networks show that our proposed approach provides better depth completions than these alternatives.

Hand detection is essential for many hand related tasks, e.g., recovering hand pose and understanding gesture. However, hand detection in uncontrolled environments is challenging due to the flexibility of wrist joint and cluttered background. We propose a convolutional neural network (CNN), which formulates in-plane rotation explicitly to solve hand detection and rotation estimation jointly. Our network architecture adopts the backbone of faster R-CNN to generate rectangular region proposals and extract local features. The rotation network takes the feature as input and estimates an in-plane rotation which manages to align the hand, if any in the proposal, to the upward direction. A derotation layer is then designed to explicitly rotate the local spatial feature map according to the rotation network and feed aligned feature map for detection. Experiments show that our method outperforms the state-of-the-art detection models on widely-used benchmarks, such as Oxford and Egohands database. Further analysis show that rotation estimation and classification can mutually benefit each other.

Access to large, diverse RGB-D datasets is critical for training RGB-D scene understanding algorithms. However, existing datasets still cover only a limited number of views or a restricted scale of spaces. In this paper, we introduce Matterport3D, a large-scale RGB-D dataset containing 10,800 panoramic views from 194,400 RGB-D images of 90 building-scale scenes. Annotations are provided with surface reconstructions, camera poses, and 2D and 3D semantic segmentations. The precise global alignment and comprehensive, diverse panoramic set of views over entire buildings enable a variety of supervised and self-supervised computer vision tasks, including keypoint matching, view overlap prediction, normal prediction from color, semantic segmentation, and region classification.

Indoor scene understanding is central to applications such as robot navigation and human companion assistance. Over the last years, data-driven deep neural networks have outperformed many traditional approaches thanks to their representation learning capabilities. One of the bottlenecks in training for better representations is the amount of available per-pixel ground truth data that is required for core scene understanding tasks such as semantic segmentation, normal prediction, and object boundary detection. To address this problem, a number of works proposed using synthetic data. However, a systematic study of how such synthetic data is generated is missing. In this work, we introduce a large-scale synthetic dataset with 500K physically-based rendered images from 45K realistic 3D indoor scenes. We study the effects of rendering methods and scene lighting on training for three computer vision tasks: surface normal prediction, semantic segmentation, and object boundary detection. This study provides insights into the best practices for training with synthetic data (more realistic rendering is worth it) and shows that pretraining with our new synthetic dataset can improve results beyond the current state of the art on all three tasks.

3D context has been shown to be extremely important for scene understanding, yet very little research has been done on integrating context information with deep neural network architectures. This paper presents an approach to embed 3D context into the topology of a neural network trained to perform holistic scene understanding. Given a depth image depicting a 3D scene, our network aligns the observed scene with a predefined 3D scene template, and then reasons about the existence and location of each object within the scene template. In doing so, our model recognizes multiple objects in a single forward pass of a 3D convolutional neural network, capturing both global scene and local object information simultaneously. To create training data for this 3D network, we generate partially synthetic depth images which are rendered by replacing real objects with a repository of CAD models of the same object category1. Extensive experiments demonstrate the effectiveness of our algorithm compared to the state of the art.

We significantly extrapolate the field of view of a photograph by learning from a roughly aligned, wide-angle guide image of the same scene category. Our method can extrapolate typical photos into complete panoramas. The extrapolation problem is formulated in the shift-map image synthesis framework. We analyze the self-similarity of the guide image to generate a set of allowable local transformations and apply them to the input image. Our guided shift-map method reserves to the scene layout of the guide image when extrapolating a photograph. While conventional shift-map methods only support translations, this is not expressive enough to characterize the self-similarity of complex scenes. Therefore we additionally allow image transformations of rotation, scaling and reflection. To handle this increase in complexity, we introduce a hierarchical graph optimization method to choose the optimal transformation at each output pixel. We demonstrate our approach on a variety of indoor, outdoor, natural, and man-made scenes.