We present a novel framework for high-fidelity novel view synthesis (NVS) from sparse images, addressing key limitations in recent feed-forward 3D Gaussian Splatting (3DGS) methods built on Vision Transformer (ViT) backbones. While ViT-based pipelines offer strong geometric priors, they are often constrained by low-resolution inputs due to computational costs. Moreover, existing generative enhancement methods tend to be 3D-agnostic, resulting in inconsistent structures across views, especially in unseen regions. To overcome these challenges, we design a Dual-Domain Detail Perception Module, which enables handling high-resolution images without being limited by the ViT backbone, and endows Gaussians with additional features to store high-frequency details. We develop a feature-guided diffusion network, which can preserve high-frequency details during the restoration process. We introduce a unified training strategy that enables joint optimization of the ViT-based geometric backbone and the diffusion-based refinement module. Experiments demonstrate that our method can maintain superior generation quality across multiple datasets.

Manipulating diverse objects with multi-fingered dexterous hands is challenging due to the high dimensionality and complex dynamics. Human-Object Interaction (HOI) datasets provide rich knowledge about task information and embodied interactions. Instead of solely imitating the human demonstrations, our method learns to holistically predict future hand-object states by leveraging these datasets. The predicted future states of the object can serve as a general-purpose reward term for reinforcement learning, reducing reliance on task-specific reward engineering and enhancing generalization across tasks. We conduct extensive experiments on three manipulation tasks in simulation and the real world. Our approach outperforms existing SOTA methods in both success rate and generalizability on novel objects. Furthermore, we validate the cross-embodiment compatibility of our methods by successfully deploying the skills on different robot hands.

Large Language Models enhanced with Retrieval Augmented Generation show strong potential in knowledge intensive tasks.However, they often encounter knowledge conflicts, where retrieved information contradicts the model's internal knowledge or exhibits internal inconsistencies.Existing methods force models into a binary choice between context and memory, leading to unreliable predictions.We argue that a more principled approach is to embrace contradictions as opportunities for deeper reasoning.To this end, we introduce Debate-of-Thoughts (DoT), a framework that transforms conflict resolution into an active deliberation process.DoT guides a single model through three phases: 1) hypothesis generation, which forms competing perspectives; 2) internal debate, where the model acts as both a proponent and a critic to stress test each view; and 3) adjudication, where the model acts as a judge to evaluate arguments based on evidence and logical consistency.We implement DoT via two complementary strategies: inference time prompt chaining and supervised fine tuning.Experiments across multiple conflict benchmarks show that DoT consistently outperforms state-of-the-art methods, while generating transparent debate transcripts that explain its decisions.By improving both accuracy and interpretability under knowledge conflicts, DoT establishes a more reliable paradigm for retrieval augmented generation systems. 1

3D scene flow represents the dense per-point motion field in dynamic scenes, playing a crucial role in various downstream tasks, including motion segmentation, dynamic scene reconstruction, 4D content generation, etc. However, previous regression-based works commonly suffer from unreliable correlations caused by locally constrained search ranges and struggle with the absence of timely feedback regarding the flow estimation uncertainty during training. To address these challenges, we propose a novel uncertainty-aware network for scene flow estimation, termed DifFlow3D, based on the conditional probabilistic diffusion model. Hierarchical diffusion-based flow estimation blocks are designed to enhance the correlation robustness and resilience to challenging cases, e.g., dynamics, noisy inputs, repetitive patterns, etc. To mitigate the generation diversity, three key flow-related features are leveraged as conditions in our diffusion model. Furthermore, we develop an uncertainty estimation module within diffusion to assess the reliability of estimated scene flow dynamically. A Hidden State Denoising strategy (HSD) is also introduced to further boost the stability of the reverse denoising process. Extensive experiments conducted on four scene flow datasets, including both synthetic and real-world datasets (FlyingThings3D, KITTI 2015, Argoverse, and Waymo Open), demonstrate the superiority of our proposed DifFlow3D. Compared to prior state-of-the-art methods, DifFlow3D has 26.0%, 36.4%, 35.3%, and 17.7% EPE3D reduction respectively across four datasets. Only trained on the synthetic FlyingThings3D dataset, our method achieves an unprecedented millimeter-level accuracy (0.0070 m EPE3D) on the real-scene KITTI dataset, highlighting its exceptional generalization capability. Additionally, our diffusion-based refinement paradigm can be seamlessly integrated as a plug-and-play module into existing scene flow networks, significantly enhancing their estimation accuracy. We also introduce our pre-trained scene flow estimator as explicit motion priors into the novel dynamic LiDAR view synthesis task, which validates its great potential for improving the 4D LiDAR reconstruction performance.

Weakly supervised 3D instance segmentation aims to reduce the high cost of point-wise annotations while maintaining competitive accuracy compared with fully supervised methods. Among various weak annotations, box annotation offers an ideal trade-off between labeling efficiency and supervision strength. However, most box-supervised methods rely on a two-stage training pipeline: 1) generating pseudo-labels, 2) training the segmentation model with pseudo-labels, which is iterative and sensitive to pseudo-label quality. To address this issue, we propose an end-to-end framework that directly learns instance masks from box annotations without an explicit pseudo-label generation and iterative relabeling and retraining stage. Specifically, we introduce a boundary-aware refinement module that adaptively learns instance boundaries from boxes through level set evolution. Furthermore, we propose a multi-scale geometric augmentation module to alleviate semantic ambiguity in overlapping regions by applying cross-view consistency constraints on predictions. Finally, we construct a multi-objective optimization framework, which improves both the training stability of level-set-evolution-based boundary refinement and the overall segmentation performance. Extensive experimental results on both indoor and outdoor datasets demonstrate that our method achieves the SOTA performance across multiple benchmark datasets with different backbones and closely approaches fully supervised counterparts.

3D Gaussian Splatting (3DGS) has recently attracted wide attentions in various areas such as 3D navigation, Virtual Reality (VR) and 3D simulation, due to its photorealistic and efficient rendering performance. High-quality reconstrution of 3DGS relies on sufficient splats and a reasonable distribution of these splats to fit real geometric surface and texture details, which turns out to be a challenging problem. We present GeoTexDensifier, a novel geometry-texture-aware densification strategy to reconstruct high-quality Gaussian splats which better comply with the geometric structure and texture richness of the scene. Specifically, our GeoTexDensifier framework carries out an auxiliary texture-aware densification method to produce a denser distribution of splats in fully textured areas, while keeping sparsity in low-texture regions to maintain the quality of Gaussian point cloud. Meanwhile, a geometry-aware splitting strategy takes depth and normal priors to guide the splitting sampling and filter out the noisy splats whose initial positions are far from the actual geometric surfaces they aim to fit, under a Validation of Depth Ratio Change checking. With the help of relative monocular depth prior, such geometry-aware validation can effectively reduce the influence of scattered Gaussians to the final rendering quality, especially in regions with weak textures or without sufficient training views. The texture-aware densification and geometry-aware splitting strategies are fully combined to obtain a set of high-quality Gaussian splats. We experiment our GeoTexDensifier framework on various datasets and compare our Novel View Synthesis results to other state-of-the-art 3DGS approaches, with detailed quantitative and qualitative evaluations to demonstrate the effectiveness of our method in producing more photorealistic 3DGS models.

White Light Interferometry (WLI) is a precise optical tool for measuring the 3D topography of microstructures. However, conventional WLI cannot capture the natural color of a sample's surface, which is essential for many microscale research applications that require both 3D geometry and color information. Previous methods have attempted to overcome this limitation by modifying WLI hardware and analysis software, but these solutions are often costly. In this work, we address this challenge from a computer vision multi-modal reconstruction perspective for the first time. We introduce OpticFusion, a novel approach that uses an additional digital optical microscope (OM) to achieve 3D reconstruction with natural color textures using multi-view WLI and OM images. Our method employs a two-step data association process to obtain the poses of WLI and OM data. By leveraging the neural implicit representation, we fuse multi-modal data and apply color decomposition technology to extract the sample's natural color. Tested on our multi-modal dataset of various microscale samples, Optic-Fusion achieves detailed 3D reconstructions with color textures. Our method provides an effective tool for practical applications across numerous microscale research fields. The source code and our real-world dataset are available at https://github.conz/zju3dv/OpticFusion.

3D Gaussian Splatting (3DGS) has shown promising performance in novel view synthesis. Previous methods adapt it to obtaining surfaces of either individual 3D objects or within limited scenes. In this paper, we make the first attempt to tackle the challenging task of large-scale scene surface reconstruction. This task is particularly difficult due to the high GPU memory consumption, different levels of details for geometric representation, and noticeable inconsistencies in appearance. To this end, we propose GigaGS, the first work for high-quality surface reconstruction for large-scale scenes using 3DGS. GigaGS first applies a partitioning strategy based on the mutual visibility of spatial regions, which effectively grouping cameras for parallel processing. To enhance the quality of the surface, we also propose novel multi-view photometric and geometric consistency constraints based on Level-of-Detail representation. In doing so, our method can reconstruct detailed surface structures. Comprehensive experiments are conducted on various datasets. The consistent improvement demonstrates the superiority of GigaGS.

Neural rendering has demonstrated remarkable success in high-quality 3D neural reconstruction and novel view synthesis with dense input views and accurate poses. However, applying it to extremely sparse, unposed views in unbounded 360° scenes remains a challenging problem. In this paper, we propose a novel neural rendering framework to accomplish the unposed and extremely sparse-view 3D reconstruction in unbounded 360° scenes. To resolve the spatial ambiguity inherent in unbounded scenes with sparse input views, we propose a layered Gaussian-based representation to effectively model the scene with distinct spatial layers. By employing a dense stereo reconstruction model to recover coarse geometry, we introduce a layer-specific bootstrap optimization to refine the noise and fill occluded regions in the reconstruction. Furthermore, we propose an iterative fusion of reconstruction and generation alongside an uncertainty-aware training approach to facilitate mutual conditioning and enhancement between these two processes. Comprehensive experiments show that our approach outperforms existing state-of-the-art methods in terms of rendering quality and surface reconstruction accuracy. Project page: https://zju3dv.github.io/free360/.

Recently, camera-based solutions have been extensively explored for scene semantic completion (SSC). Despite their success in visible areas, existing methods struggle to capture complete scene semantics due to frequent visual occlusions. To address this limitation, this paper presents the first satellite-ground cooperative SSC framework, i.e., SGFormer, exploring the potential of satellite-ground image pairs in the SSC task. Specifically, we propose a dual-branch architecture that encodes orthogonal satellite and ground views in parallel, unifying them into a common domain. Additionally, we design a ground-view guidance strategy that corrects satellite image biases during feature encoding, addressing misalignment between satellite and ground views. Moreover, we develop an adaptive weighting strategy that balances contributions from satellite and ground views. Experiments demonstrate that SG-Former outperforms the state of the art on SemanticKITTI and SSCBench-KITTI-360 datasets. Our code is available on https://github.com/gxytcrc/SGFormer.

This paper presents a novel camera relocalization method, STDLoc, which leverages Feature Gaussian as scene representation. STDLoc is a full relocalization pipeline that can achieve accurate relocalization without relying on any pose prior. Unlike previous coarse-to-fine localization methods that require image retrieval first and then feature matching, we propose a novel sparse-to-dense localization paradigm. Based on this scene representation, we introduce a novel matching-oriented Gaussian sampling strategy and a scene-specific detector to achieve efficient and robust initial pose estimation. Furthermore, based on the initial localization results, we align the query feature map to the Gaussian feature field by dense feature matching to enable accurate localization. The experiments on indoor and outdoor datasets show that STDLoc outperforms current state-of-the-art localization methods in terms of localization accuracy and recall. Our code is available on the project website: https://zju3dv.github.io/STDLoc.

In this paper, we introduce a novel, truly prior-free 3D object tracking method that operates without given any model or training priors. Unlike existing methods that typically require pre-defined 3D models or specific training datasets as priors, which limit their applicability, our method is free from these constraints. Our method consists of a geometry generation module and a pose optimization module. Its core idea is to enable these two modules to automatically and iteratively enhance each other, thereby gradually building all the necessary information for the tracking task. We thus call the method as Bidirectional Iterative Tracking(BIT). The geometry generation module starts without priors and gradually generates high-precision mesh models for tracking, while the pose optimization module generates additional data during object tracking to further refine the generated models. Moreover, the generated 3D models can be stored and easily reused, allowing for seamless integration into various other tracking systems, not just our methods. Experimental results demonstrate that BIT outperforms many existing methods, even those that extensively utilize prior knowledge, while BIT does not rely on such information. Additionally, the generated 3D models deliver results comparable to actual 3D models, highlighting their superior and innovative qualities. The code is available at https://github.com/songxiuqiang/BIT.git.

Recently, 3D Gaussian Splatting (3DGS) has shown encouraging performance for open vocabulary scene understanding tasks. However, previous methods cannot distinguish 3D instance-level information, which usually predicts a heatmap between the scene feature and text query. In this paper, we propose PanoGS, a novel and effective 3D panoptic open vocabulary scene understanding approach. Technically, to learn accurate 3D language features that can scale to large indoor scenarios, we adopt the pyramid tri-plane to model the latent continuous parametric feature space and use a 3D feature decoder to regress the multi-view fused 2D feature cloud. Besides, we propose language-guided graph cuts that synergistically leverage reconstructed geometry and learned language cues to group 3D Gaussian primitives into a set of super-primitives. To obtain 3D consistent instance, we perform graph clustering based segmentation with SAM-guided edge affinity computation between different super-primitives. Extensive experiments on widely used datasets show better or more competitive performance on 3D panoptic open vocabulary scene understanding. Project page: https://zju3dv.github.io/panogs.

Recent advances in large reconstruction and generative models have significantly improved scene reconstruction and novel view generation. However, due to compute limitations, each inference with these large models is confined to a small area, making long-range consistent scene generation challenging. To address this, we propose StarGen, a novel framework that employs a pre-trained video diffusion model in an autoregressive manner for long-range scene generation. The generation of each video clip is conditioned on the 3D warping of spatially adjacent images and the tem-porally overlapping image from previously generated clips, improving spatiotemporal consistency in long-range scene generation with precise pose control. The spatiotemporal condition is compatible with various input conditions, facilitating diverse tasks, including sparse view interpolation, perpetual view generation, and layout-conditioned city generation. Quantitative and qualitative evaluations demonstrate StarGen’s superior scalability, fidelity, and pose accuracy compared to state-of-the-art methods. Project page: https://zju3dv.github.io/StarGen.

Event cameras, known for their high temporal resolution and ability to capture asynchronous changes, have gained significant attention for their potential in feature tracking, especially in challenging conditions. However, event cameras lack the fine-grained texture information that conventional cameras provide, leading to error accumulation in tracking. To address this, we propose a novel framework, BlinkTrack, which integrates event data with grayscale images for high-frequency feature tracking. Our method extends the traditional Kalman filter into a learning-based framework, utilizing differentiable Kalman filters in both event and image branches. This approach improves single-modality tracking and effectively solves the data association and fusion from asynchronous event and image data. We also introduce new synthetic and augmented datasets to better evaluate our model. Experimental results indicate that BlinkTrack significantly outperforms existing methods, exceeding 80 FPS with multi-modality data and 100 FPS with preprocessed event data. Codes and dataset are available at https://github.com/ColieShen/BlinkTrack.

Neural implicit reconstruction via volume rendering has demonstrated its effectiveness in recovering dense 3D surfaces. However, it is non-trivial to simultaneously recover meticulous geometry and preserve smoothness across regions with differing characteristics. To address this issue, previous methods typically employ geometric priors, which are often constrained by the performance of the prior models. In this paper, we propose ND-SDF, which learns a Normal Deflection field to represent the angular deviation between the scene normal and the prior normal. Unlike previous methods that uniformly apply geometric priors on all samples, introducing significant bias in accuracy, our proposed normal deflection field dynamically learns and adapts the utilization of samples based on their specific characteristics, thereby improving both the accuracy and effectiveness of the model. Our method not only obtains smooth weakly textured regions such as walls and floors but also preserves the geometric details of complex structures. In addition, we introduce a novel ray sampling strategy based on the deflection angle to facilitate the unbiased rendering process, which significantly improves the quality and accuracy of intricate surfaces, especially on thin structures. Consistent improvements on various challenging datasets demonstrate the superiority of our method.

Multi-session visual SLAM systems enable 6-DoF camera localization along with long-term maintenance and expansion of the global map, by utilizing image data from different sessions. However, in large-scale environments, these systems often suffer from severe scale drift. While modern SLAM systems attempt to maintain global map consistency through loop detection and correction, they still face challenges in terms of convergence and accuracy. In this paper, we propose a robust large-scale multi-session SLAM system for long-term localization and mapping that achieves global consistency. Furthermore, to address the backend optimization problem in large-scale environments, we introduce a hierarchical optimization strategy based on the graph structure. More specifically, a subgraph structure is introduced to reduce the size of problem while effectively propagating scale correction information. In addition, a hierarchical strategy enables coarse-to-fine updates of the graph states. Experimental results not only demonstrate that our method efficiently optimizes the pose graph and maintains map consistency in large-scale environments, but also highlight the effectiveness and scalability of the proposed approach.

Recently, neural radiance fields (NeRF) have gained significant attention in the field of visual localization. However, existing NeRF-based approaches either lack geometric constraints or require extensive storage for feature matching, limiting their practical applications. To address these challenges, we propose an efficient and novel visual localization approach based on the neural implicit map with complementary features. Specifically, to enforce geometric constraints and reduce storage requirements, we implicitly learn a 3D keypoint descriptor field, avoiding the need to explicitly store point-wise features. To further address the semantic ambiguity of descriptors, we introduce additional semantic contextual feature fields, which enhance the quality and reliability of 2D-3D correspondences. Besides, we propose descriptor similarity distribution alignment to minimize the domain gap between 2D and 3D feature spaces during matching. Finally, we construct the matching graph using both complementary descriptors and contextual features to establish accurate 2D3D correspondences for 6-DoF pose estimation. Compared with the recent NeRF-based approaches, our method achieves a$3 \times$faster training speed and a$45 \times$reduction in model storage. Extensive experiments on two widely used datasets demonstrate that our approach outperforms or is highly competitive with other state-of-the-art NeRF-based visual localization methods. Project page: https://zju3dv.github.io/neuraloc

Visual-inertial odometry (VIO) has made significant progress in various applications. However, one of the key challenges in VIO is the efficient and robust fusion of visual and inertial measurements, particularly while mitigating the impact of sensor failures. To address this challenge, we propose a new learning-based VIO system, i.e., DW-VIO, which is able to integrate multiple sensors and provide robust state estimations. To this end, we design a novel deep learning-based data-fusion approach that dynamically associates information from multiple sensors to predict sensor weights for optimization. Moreover, in order to improve the efficiency, we present several real-time optimization techniques including a fast patch graph constructor and an efficient GPU-accelerated multi-factor bundle adjustment layer. Experimental results show that DW-VIO outperforms most state-of-the-art (SOTA) methods on the EuRoC MAV, ETH3D-SLAM, and KITTI-360 benchmarks across various challenging sequences. Additionally, it maintains a minimum of 20 frames per second (FPS) on a single RTX 3060 GPU with high-resolution input, highlighting its efficiency.

Monocular depth estimation is essential for applications such as autonomous navigation and 3D reconstruction. However, achieving accurate and temporally consistent depth estimation in dynamic environments remains challenging due to scale ambiguity, sensitivity to dynamic objects, and inconsistent depth predictions. Traditional SLAM-based methods ensure global consistency but perform poorly in dynamic scenes, while deep learning-based approaches suffer from the absence of absolute scale and temporal stability. To address these issues, we propose a Real-Time Consistent Monocular Depth Recovery System that combines ORB-SLAM3 for sparse depth initialization, a ViT-based depth completion network, and a motion segmentation module to improve robustness in dynamic environments. Additionally, we introduce a dual-weight fusion module that adaptively balances RGB semantic features and geometric depth priors, ensuring high accuracy and consistency. Our system jointly optimizes both static and dynamic regions to produce globally scale-consistent dense depth maps with improved temporal stability. Extensive experiments on benchmark datasets demonstrate that our approach outperforms existing methods in terms of depth accuracy, temporal consistency, and robustness in dynamic scenes, while maintaining real-time performance.

Recent feature matching approaches like ETO have focused on developing lightweight matching algorithms for real-time applications. However, their lack of cross-image feature interaction and sufficient refinement often lead to a decline in matching accuracy. To address these challenges, we propose ETO+, a novel and accurate feature matching algorithm that incorporates a lightweight yet efficient bidirectional interaction module and multi-stage refinement. Specifically, we introduce Trans-CNN, a bidirectional feature interaction module that integrates CNN- and transformer-based techniques to enhance both intra-image feature refinement and inter-image feature fusion, all while maintaining a comparable computational cost. Furthermore, by leveraging the inherent sparsity of local feature matching, we propose an efficient strategy to adaptively reallocate computational resources within the network. Additionally, we design an adaptive loss function that mitigates the impact of large matching errors, thereby improving overall robustness. Extensive experiments on widely used datasets demonstrate that our approach achieves a strong balance between accuracy and computational efficiency. It outperforms ETO by 7.9 in AUC@5 on MegaDepth, respectively, while being about 40% faster than E-LoFTR.

Recent advances in 3D Gaussian Splatting (3DGS) have enabled real-time, high-fidelity view synthesis, but remain critically dependent on camera poses estimated by Structure-from-Motion (SfM), which is notoriously unreliable in textureless indoor environments. To eliminate this dependency, recent pose-free variants have been proposed, yet they often fail under abrupt camera motion due to unstable initialization and purely photometric objectives. In this work, we introduce **Nope-RoomGS**, an optimization framework with no need for camera pose inputs, which effectively addresses the textureless regions and abrupt camera motion in indoor room environments through a local-to-global optimization paradigm for 3DGS reconstruction. In the local stage, we propose a lightweight local neural geometric representation to bootstrap a set of reliable local 3D Gaussians for separated short video clips, regularized by multi-frame tracking constraints and foundation model depth priors. This enables reliable initialization even in textureless regions or under abrupt camera motions. In the global stage, we fuse local 3D Gaussians into a unified 3DGS representation through an alternating optimization strategy that jointly refines camera poses and Gaussian parameters, effectively mitigating gradient interference between them. Furthermore, we decompose camera pose optimization based on a piecewise planarity assumption, further enhancing robustness under abrupt camera motion. Extensive experiments on Replica, ScanNet and Tanks & Temples demonstrate the state-of-the-art performance of our method in both camera pose estimation and novel view synthesis.

3D reconstruction of indoor and urban environments is a prominent research topic with various downstream applications. However, existing geometric priors for addressing low-texture regions in indoor and urban settings often lack global consistency. Moreover, Gaussian Splatting and implicit SDF fields often suffer from discontinuities or exhibit computational inefficiencies, resulting in a loss of detail. To address these issues, we propose an Atlanta-world guided implicit-structured Gaussian Splatting that achieves smooth indoor and urban scene reconstruction while preserving high-frequency details and rendering efficiency. By leveraging the Atlanta-world model, we ensure the accurate surface reconstruction for low-texture regions, while the proposed novel implicit-structured GS representations provide smoothness without sacrificing efficiency and high-frequency details. Specifically, we propose a semantic GS representation to predict the probability of all semantic regions and deploy a structure plane regularization with learnable plane indicators for global accurate surface reconstruction. Extensive experiments demonstrate that our method outperforms state-of-the-art approaches in both indoor and urban scenes, delivering superior surface reconstruction quality.

This paper presents GS-RoadPatching, an inpainting method for driving scene completion by referring to completely reconstructed regions, which are represented by 3D Gaussian Splatting (3DGS). Unlike existing 3DGS inpainting methods that perform generative completion relying on 2D perspective-view-based diffusion or GAN models to predict limited appearance or depth cues for missing regions, our approach enables substitutional scene inpainting and editing directly through the 3DGS modality, extricating it from requiring spatial-temporal consistency of 2D cross-modals and eliminating the need for time-intensive retraining of Gaussians. Our key insight is that the highly repetitive patterns in driving scenes often share multi-modal similarities within the implicit 3DGS feature space and are particularly suitable for structural matching to enable effective 3DGS-based substitutional inpainting. Practically, we construct feature-embedded 3DGS scenes to incorporate a patch measurement method for abstracting local context at different scales and, subsequently, propose a structural search method to find candidate patches in 3D space effectively. Finally, we propose a simple yet effective substitution-and-fusion optimization for better visual harmony. We conduct extensive experiments on multiple publicly available datasets to demonstrate the effectiveness and efficiency of our proposed method in driving scenes, and the results validate that our method achieves state-of-the-art performance compared to the baseline methods in terms of both quality and interoperability. Additional experiments in general scenes also demonstrate the applicability of the proposed 3D inpainting strategy. The project page and code are available at: https://shanzhaguoo.github.io/GS-RoadPatching/.

Real-time 3D reconstruction is a fundamental task in computer graphics. Recently, differentiable-rendering-based SLAM system has demonstrated significant potential, enabling photorealistic scene rendering through learnable scene representations such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). Current differentiable rendering methods face dual challenges in real-time computation and sensor noise sensitivity, leading to degraded geometric fidelity in scene reconstruction and limited practicality. To address these challenges, we propose a novel real-time system EGG-Fusion, featuring robust sparse-to-dense camera tracking and a geometry-aware Gaussian surfel mapping module, introducing an information filter-based fusion method that explicitly accounts for sensor noise to achieve high-precision surface reconstruction. The proposed differentiable Gaussian surfel mapping effectively models multi-view consistent surfaces while enabling efficient parameter optimization. Extensive experimental results demonstrate that the proposed system achieves a surface reconstruction error of 0.6cm on standardized benchmark datasets including Replica and ScanNet++, representing over 20% improvement in accuracy compared to state-of-the-art (SOTA) GS-based methods. Notably, the system maintains real-time processing capabilities at 24 FPS, establishing it as one of the most accurate differentiable-rendering-based real-time reconstruction systems. Project Page: https://zju3dv.github.io/eggfusion/.

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.

Reconstructing urban street scenes is crucial due to its vital role in applications such as autonomous driving and urban planning. These scenes are characterized by long, narrow camera trajectories, occlusion, complex object relationships, and sparse data across multiple scales. Despite recent advancements, existing surface reconstruction methods, which are primarily designed for object-centric scenarios, struggle to adapt effectively to the unique characteristics of street scenes. To address this challenge, we introduce StreetSurfGS, the first method to employ Gaussian Splatting specifically tailored for scalable urban street scene surface reconstruction. StreetSurfGS utilizes a planar-based octree representation and segmented training to reduce memory costs, accommodate unique camera characteristics, and improve scalability. Additionally, to mitigate depth inaccuracies caused by object overlap, we propose a guided smoothing strategy within regularization to eliminate inaccurate boundary points and outliers. Furthermore, to address sparse views and multi-scale challenges, we use a dual-step matching strategy that leverages adjacent and long-term information. Extensive experiments validate the efficacy of StreetSurfGS in both novel view synthesis and surface reconstruction.

6D object tracking plays an important role in various applications, including robotic manipulation and virtual reality. While current methodologies have achieved significant advancements through the use of CAD models, multi-modal sensor data, and category-level assumptions, such resources are often inaccessible in open-world scenarios. Consequently, tracking 6D object poses using only RGB data in such scenarios remains a challenging task. In this paper, we introduce Zero6DOT, an innovative and efficient method for real-time tracking of unknown 6D object poses in monocular RGB video sequences at 8Hz. Our approach requires only the mask of the initial frame, eliminating the need for additional data. The core of Zero6DOT lies in its ability to establish high-quality correspondences across images, from which accurate poses are derived. To achieve this, we employ a transformer-based neural network to predict initial long-term correspondences across frames and integrate a robust Dynamic Units System to refine these predictions. This combination facilitates precise pose tracking while maintaining both efficiency and robustness, even under challenging conditions such as object disappearance, reappearance, and handheld motion. The effectiveness of our approach has been rigorously evaluated through both qualitative and quantitative analyses on the OnePose, YCB-V, and RBOT datasets. The results demonstrate the potential of our proposed Zero6DOT to redefine 6D object pose tracking for real-world scenarios.

Capturing the human body and clothing from videos has obtained significant progress in recent years, but several challenges remain to be addressed. Previous methods reconstruct the 3D bodies and garments from videos with self-rotating human motions or capture the body and clothing separately based on neural implicit fields. However, the reconstruction methods for self-rotating motions may cause instable tracking on dynamic videos with arbitrary human motions, while implicit fields based methods are limited to inefficient rendering and low quality synthesis. To solve these problems, we propose a new method, called CloCap-GS, for clothed human performance capture with 3D Gaussian Splatting. Specifically, we align 3D Gaussians with the deforming geometries of body and clothing, and leverage photometric constraints formed by matching Gaussians renderings with input video frames to recover temporal deformations of the dense template geometry. The geometry deformations and Gaussians properties of both the body and clothing are optimized jointly, achieving both dense geometry tracking and novel-view synthesis. In addition, we introduce a physics-aware material-varying cloth model to preserve physically-plausible cloth dynamics and body-clothing interactions that is pre-trained in a self-supervised manner without preparing training data. Compared with the existing methods, our method improves the accuracy of dense geometry tracking and quality of novel-view synthesis for a variety of daily garment types (e.g., loose clothes). Extensive experiments in both quantitative and qualitative evaluations demonstrate the effectiveness of CloCap-GS on real sparse-view or monocular videos.

Recently, 3D Gaussian Splatting (3DGS) has attracted widespread attention due to its high-quality rendering, and ultra-fast training and rendering speed. However, due to the unstructured and irregular nature of Gaussian point clouds, it is difficult to guarantee geometric reconstruction accuracy and multi-view consistency simply by relying on image reconstruction loss. Although many studies on surface reconstruction based on 3DGS have emerged recently, the quality of their meshes is generally unsatisfactory. To address this problem, we propose a fast planar-based Gaussian splatting reconstruction representation (PGSR) to achieve high-fidelity surface reconstruction while ensuring high-quality rendering. Specifically, we first introduce an unbiased depth rendering method, which directly renders the distance from the camera origin to the Gaussian plane and the corresponding normal map based on the Gaussian distribution of the point cloud, and divides the two to obtain the unbiased depth. We then introduce single-view geometric, multi-view photometric, and geometric regularization to preserve global geometric accuracy. We also propose a camera exposure compensation model to cope with scenes with large illumination variations. Experiments on indoor and outdoor scenes show that the proposed method achieves fast training and rendering while maintaining high-fidelity rendering and geometric reconstruction, outperforming 3DGS-based and NeRF-based methods.

Dense SLAM based on monocular cameras does indeed have immense application value in the field of AR/VR, especially when it is performed on a mobile device. In this article, we propose a novel method that integrates a light-weight depth completion network into a sparse SLAM system using a multi-basis depth representation, so that dense mapping can be performed online even on a mobile phone. Specifically, we present a specifically optimized multi-basis depth completion network, called BBC-Net, tailored to the characteristics of traditional sparse SLAM systems. BBC-Net can predict multiple balanced bases and a confidence map from a monocular image with sparse points generated by off-the-shelf keypoint-based SLAM systems. The final depth is a linear combination of predicted depth bases that can be easily optimized by tuning the corresponding weights. To seamlessly incorporate the weights into traditional SLAM optimization and ensure efficiency and robustness, we design a set of depth weight factors, which makes our network a versatile plug-in module, facilitating easy integration into various existing sparse SLAM systems and significantly enhancing global depth consistency through bundle adjustment. To verify the portability of our method, we integrate BBC-Net into two representative SLAM systems. The experimental results on various datasets show that the proposed method achieves better performance in monocular dense mapping than the state-of-the-art methods. We provide an online demo running on a mobile phone, which verifies the efficiency and mapping quality of the proposed method in real-world scenarios.

Visual localization plays an important role in the applications of Augmented Reality (AR), which enable AR devices to obtain their 6-DoF pose in the pre-build map in order to render virtual content in real scenes. However, most existing approaches can not perform novel view rendering and require large storage capacities for maps. To overcome these limitations, we propose an efficient visual localization method capable of high-quality rendering with fewer parameters. Specifically, our approach leverages 3D Gaussian primitives as the scene representation. To ensure precise 2D-3D correspondences for pose estimation, we develop an unbiased 3D scene-specific descriptor decoder for Gaussian primitives, distilled from a constructed feature volume. Additionally, we introduce a salient 3D landmark selection algorithm that selects a suitable primitive subset based on the saliency score for localization. We further regularize key Gaussian primitives to prevent anisotropic effects, which also improves localization performance. Extensive experiments on two widely used datasets demonstrate that our method achieves superior or comparable rendering and localization performance to state-of-the-art implicit-based visual localization approaches. Code and data are available at project page: https://zju3dv.github.io/splatloc.

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/.

We present DiffInDScene, a novel framework for tack-ling the problem of high-quality 3D indoor scene generation, which is challenging due to the complexity and diversity of the indoor scene geometry. Although diffusion-based generative models have previously demonstrated impressive performance in image generation and object-level 3D generation, they have not yet been applied to room-level 3D generation due to their computationally intensive costs. In DiffInDScene, we propose a cascaded 3D diffusion pipeline that is efficient and possesses strong generative performance for Truncated Signed Distance Function (TSDF). The whole pipeline is designed to run on a sparse occupancy space in a coarse-to-fine fashion. Inspired by KinectFusion's incremental alignment and fusion of local TSDF volumes, we propose a diffusion-based SDF fusion approach that iteratively diffuses andfuses local TSDF volumes, facilitating the generation of an entire room environment. The generated results demonstrate that our work is capable to achieve high-quality room generation directly in three-dimensional space, starting from scratch. In addition to the scene generation, the final part of DiffInDScene can be used as a post-processing module to refine the 3D reconstruction results from multi-view stereo. According to the user study, the mesh quality generated by our DifflnD-Scene can even outperform the ground truth mesh provided by ScanNet.

Scene flow estimation, which aims to predict per-point 3D displacements of dynamic scenes, is a fundamen-tal task in the computer vision field. However, previ-ous works commonly suffer from unreliable correlation caused by locally constrained searching ranges, and struggle with accumulated inaccuracy arising from the coarse-to-fine structure. To alleviate these problems, we propose a novel uncertainty-aware scene flow estimation network(DifFlow3D) with the diffusion probabilistic model. Iter-ative diffusion-based refinement is designed to enhance the correlation robustness and resilience to challenging cases, e.g. dynamics, noisy inputs, repetitive patterns, etc. To re-strain the generation diversity, three key flow-related features are leveraged as conditions in our diffusion model. Furthermore, we also develop an uncertainty estimation module within diffusion to evaluate the reliability of esti-mated scene flow. Our DifFlow3D achieves state-of-the-art performance, with 24.0% and 29.1% EPE3D reduction respectively on FlyingThings3D and KITTI 2015 datasets. Notably, our method achieves an unprecedented millimeter-level accuracy (O.0078m in EPE3D) on the KITTI dataset. Additionally, our diffusion-based refinement paradigm can be readily integrated as a plug-and-play module into ex-isting scene flow networks, significantly increasing their estimation accuracy. Codes are released at https:// github.com/IRMVLab/DifFlow3D.

One of the key challenges of visual Simultaneous Localization and Mapping (SLAM) in large-scale environments is how to effectively use global localization to correct the cumulative errors from long-term tracking. This challenge presents itself in two main aspects: first, the difficulty for robots in revisiting previous locations to perform loop closure, and second, the considerable memory resources required to maintain point-cloud-based global maps. Recent solutions have resorted into neural networks, using satellite images as the references for ground-level localization. However, most of these methods merely provide cross-view patch-matching results, which leads to unfeasible in integration with the SLAM system. To address these issues, we present a semantic-based cross-view localization method. This approach combines semantic information with a reward and penalty mechanism, enabling us to obtain a global probability map and achieve precise 3-degree-of-freedom (3-DoF) localization. Based on that, we develop a SLAM system that capitalizes on satellite imagery for global localization. This strategy effectively bridges the gap between SLAM and real-world coordinates while also substantially reducing accumulated errors. Our experimental results demonstrate that our global localization method significantly outperforms existing satellite-based systems. Moreover, in scenarios where the robot struggles to find loop closures, employing our localization method improves the SLAM accuracy.

We propose a real-time visual-inertial dense SLAM system that utilizes the online data streams from back-to-back dual fisheye cameras setup, providing 360◦coverage of the environment. Firstly, we employ a sliding-window-based front-end to estimate real-time poses from the binocular fisheye images and IMU data. Then, we implement a lightweight panoramic depth completion network based on multi-basis depth representation. The network takes panoramic images (obtained by stitching dual-fisheye images with extrinsics and intrinsic parameters) and sparse depths (generated by the front-end local tracking) as input and predicts multiple depth bases along with corresponding confidence as output. The final dense depth is the linear combination of the multiple depth bases. Thanks to the multi-basis depth representation, we can continuously optimize the 360° depth with the traditional optimizer to achieve higher global consistency in depth. We conducted experiments on both simulated and real-world datasets to evaluate our method. The results demonstrate that the proposed method outperforms SoTA methods in terms of depth prediction and 3D reconstruction. In addition, we develop a demo that can run on a mobile to demonstrate the real-time capabilities of our method.

In transformer-based methods for point cloud instance segmentation, bipartite matching is used to establish one-to-one correspondences between predictions and ground truths. However, in early training stages, matches can be unstable and inconsistent between epochs, requiring the model to frequently adjust its learning path, thus reducing the quality of model convergence. To address this challenge, we propose the contrastive mask denoising transformer for 3D instance segmentation, which utilizes a mask denoising module to guide the model towards a more stable optimization path in early training stages. Furthermore, we introduce a multi-pattern-aware query selection module to assist the model learn multiple patterns at one position such that clustered objects can be discerned. In addition, the proposed modules are "plug and play", which can easily be integrated into transformer-based architectures. Experimental results on ScanNetv2 dataset show that the proposed modules improve the performance of multiple pipelines, notably achieving +1.0 mAP on the main pipeline.

In this paper, we propose a novel multi-view stereo (MVS) framework that gets rid of the depth range prior. Unlike recent prior-free MVS methods that work in a pair-wise manner, our method simultaneously considers all the source images. Specifically, we introduce a Multi-view Disparity Attention (MDA) module to aggregate long-range context information within and across multi-view images. Considering the asymmetry of the epipolar disparity flow, the key to our method lies in accurately modeling multi-view geometric constraints. We integrate pose embedding to encapsulate information such as multi-view camera poses, providing implicit geometric constraints for multi-view disparity feature fusion dominated by attention. Additionally, we construct corresponding hidden states for each source image due to significant differences in the observation quality of the same pixel in the reference frame across multiple source frames. We explicitly estimate the quality of the current pixel corresponding to sampled points on the epipolar line of the source image and dynamically update hidden states through the uncertainty estimation module. Extensive results on the DTU dataset and Tanks\&Temple benchmark demonstrate the effectiveness of our method.

We tackle the efficiency problem of learning local feature matching.Recent advancements have given rise to purely CNN-based and transformer-based approaches, each augmented with deep learning techniques. While CNN-based methods often excel in matching speed, transformer-based methods tend to provide more accurate matches. We propose an efficient transformer-based network architecture for local feature matching.This technique is built on constructing multiple homography hypotheses to approximate the continuous correspondence in the real world and uni-directional cross-attention to accelerate the refinement. On the YFCC100M dataset, our matching accuracy is competitive with LoFTR, a state-of-the-art transformer-based architecture, while the inference speed is boosted to 4 times, even outperforming the CNN-based methods.Comprehensive evaluations on other open datasets such as Megadepth, ScanNet, and HPatches demonstrate our method's efficacy, highlighting its potential to significantly enhance a wide array of downstream applications.

Signed Distance Function (SDF)-based volume rendering has demonstrated significant capabilities in surface reconstruction. Although promising, SDF-based methods often fail to capture detailed geometric structures, resulting in visible defects. By comparing SDF-based volume rendering to density-based volume rendering, we identify two main factors within the SDF-based approach that degrade surface quality: SDF-to-density representation and geometric regularization. These factors introduce challenges that hinder the optimization of the SDF field. To address these issues, we introduce NeuRodin, a novel two-stage neural surface reconstruction framework that not only achieves high-fidelity surface reconstruction but also retains the flexible optimization characteristics of density-based methods. NeuRodin incorporates innovative strategies that facilitate transformation of arbitrary topologies and reduce artifacts associated with density bias. Extensive evaluations on the Tanks and Temples and ScanNet++ datasets demonstrate the superiority of NeuRodin, showing strong reconstruction capabilities for both indoor and outdoor environments using solely posed RGB captures. Project website: https://open3dvlab.github.io/NeuRodin/

Diffusion-based methods have achieved remarkable achievements in 2D image or 3D object generation, however, the generation of 3D scenes and even $360^{\circ}$ images remains constrained, due to the limited number of scene datasets, the complexity of 3D scenes themselves, and the difficulty of generating consistent multi-view images. To address these issues, we first establish a large-scale panoramic video-text dataset containing millions of consecutive panoramic keyframes with corresponding panoramic depths, camera poses, and text descriptions. Then, we propose a novel text-driven panoramic generation framework, termed DiffPano, to achieve scalable, consistent, and diverse panoramic scene generation. Specifically, benefiting from the powerful generative capabilities of stable diffusion, we fine-tune a single-view text-to-panorama diffusion model with LoRA on the established panoramic video-text dataset. We further design a spherical epipolar-aware multi-view diffusion model to ensure the multi-view consistency of the generated panoramic images. Extensive experiments demonstrate that DiffPano can generate scalable, consistent, and diverse panoramic images with given unseen text descriptions and camera poses.

6-DoF object pose estimation from a monocular image is a challenging problem, where a post-refinement procedure is generally needed for high-precision estimation. In this paper, we propose a framework, dubbed RNNPose, based on a recurrent neural network (RNN) for object pose refinement, which is robust to erroneous initial poses and occlusions. During the recurrent iterations, object pose refinement is formulated as a non-linear least squares problem based on the estimated correspondence field (between a rendered image and the observed image). The problem is then solved by a differentiable Levenberg-Marquardt (LM) algorithm enabling end-to-end training. The correspondence field estimation and pose refinement are conducted alternately in each iteration to improve the object poses. Furthermore, to improve the robustness against occlusion, we introduce a consistency-check mechanism based on the learned descriptors of the 3D model and observed 2D images, which downweights the unreliable correspondences during pose optimization. We evaluate RNNPose on several public datasets, including LINEMOD, Occlusion-LINEMOD, YCB-Video and TLESS. We demonstrate state-of-the-art performance and strong robustness against severe clutter and occlusion in the scenes. Extensive experiments validate the effectiveness of our proposed method. Besides, the extended system based on RNNPose successfully generalizes to multi-instance scenarios and achieves top-tier performance on the TLESS dataset.

This paper addresses the challenge of reconstructing 3D indoor scenes from multi-view images. Many previous works have shown impressive reconstruction results on textured objects, but they still have difficulty in handling low-textured planar regions, which are common in indoor scenes. An approach to solving this issue is to incorporate planar constraints into the depth map estimation in multi-view stereo-based methods, but the per-view plane estimation and depth optimization lack both efficiency and multi-view consistency. In this work, we show that the planar constraints can be conveniently integrated into the recent implicit neural representation-based reconstruction methods. Specifically, we use an MLP network to represent the signed distance function as the scene geometry. Based on the Manhattan-world assumption and the Atlanta-world assumption, planar constraints are employed to regularize the geometry in floor and wall regions predicted by a 2D semantic segmentation network. To resolve the inaccurate segmentation, we encode the semantics of 3D points with another MLP and design a novel loss that jointly optimizes the scene geometry and semantics in 3D space. Experiments on ScanNet and 7-Scenes datasets show that the proposed method outperforms previous methods by a large margin on 3D reconstruction quality.

Nowadays the application of AR is expanding from small or medium environments to large-scale environments, where the visual-based localization in the large-scale environments becomes a critical demand. Current visual-based localization techniques face robustness challenges in complex large-scale environments, requiring tremendous number of data with groundtruth localization for algorithm benchmarking or model training. The previous groundtruth solutions can only be used outdoors, or require high equipment/labor costs, so they cannot be scalable to large environments for both indoors and outdoors, nor can they produce large amounts of data at a feasible cost. In this work, we propose LSFB, a novel low-cost and scalable framework to build localization benchmark in large-scale indoor and outdoor environments. The key is to reconstruct an accurate HD map of the environment. For each visual-inertial sequence captured in the environment, the groundtruth poses are obtained by joint optimization taking both the HD map and visual-inertial constraints. The experiments demonstrate the obtained groundtruth poses have cm-level accuracy. We use the proposed method to collect a localization dataset by mobile phones and AR glasses in various environments with various motions, and release the dataset as the first large-scale localization benchmark for AR.

Structure-from-Motion is a technology used to obtain scene structure through image collection, which is a fundamental problem in computer vision. For unordered Internet images, SfM is very slow due to the lack of prior knowledge about image overlap. For sequential images, knowing the large overlap between adjacent frames, SfM can adopt a variety of acceleration strategies, which are only applicable to sequential data. To further improve the reconstruction efficiency and break the gap of strategies between these two kinds of data, this paper presents an efficient covisibility-based incremental SfM. Different from previous methods, we exploit covisibility and registration dependency to describe the image connection which is suitable to any kind of data. Based on this general image connection, we propose a unified framework to efficiently reconstruct sequential images, unordered images, and the mixture of these two. Experiments on the unordered images and mixed data verify the effectiveness of the proposed method, which is three times faster than the state-of-the-art on feature matching, and an order of magnitude faster on reconstruction without sacrificing the accuracy. The source code is publicly available athttps://github.com/openxrlab/xrsfm.

The delay of rendering on AR devices requires prediction of head motion using sensor data acquired tens of even one hundred milliseconds ago to avoid misalignment between the virtual content and the physical world, where the misalignment will lead to a sense of time latency and dizziness for users. To solve the problem, we propose a method for the 6DoF motion prediction to compensate for the time latency. Compared with traditional hand-crafted methods, our method is based on deep learning, which has better motion prediction ability to deal with complex human motion. In particular, we propose a MOtion UNcerTainty encode decode network (MOUNT) that estimates the uncertainty of input data and predicts the uncertainty of output motion to improve the prediction accuracy and smoothness. Experiments on the EuRoC and our collected dataset demonstrate that our method significantly outperforms the traditional method and greatly improves AR visual effects.

It is typically challenging for visual or visual-inertial odometry systems to handle the problems of dynamic scenes and pure rotation. In this work, we design a novel visual-inertial odometry (VIO) system called RD-VIO to handle both of these two problems. First, we propose an IMU-PARSAC algorithm which can robustly detect and match keypoints in a two-stage process. In the first state, landmarks are matched with new keypoints using visual and IMU measurements. We collect statistical information from the matching and then guide the intra-keypoint matching in the second stage. Second, to handle the problem of pure rotation, we detect the motion type and adapt the deferred-triangulation technique during the data-association process. We make the pure-rotational frames into the special subframes. When solving the visual-inertial bundle adjustment, they provide additional constraints to the pure-rotational motion. We evaluate the proposed VIO system on public datasets and online comparison. Experiments show the proposed RD-VIO has obvious advantages over other methods in dynamic environments.

Virtual content creation and interaction play an important role in modern 3D applications. Recovering detailed 3D models from real scenes can significantly expand the scope of its applications and has been studied for decades in the computer vision and computer graphics community. In this work, we propose Vox-Surf, a voxel-based implicit surface representation. Our Vox-Surf divides the space into finite sparse voxels, where each voxel is a basic geometry unit that stores geometry and appearance information on its corner vertices. Due to the sparsity inherited from the voxel representation, Vox-Surf is suitable for almost any scene and can be easily trained end-to-end from multiple view images. We utilize a progressive training process to gradually cull out empty voxels and keep only valid voxels for further optimization, which greatly reduces the number of sample points and improves inference speed. Experiments show that our Vox-Surf representation can learn fine surface details and accurate colors with less memory and faster rendering than previous methods. The resulting fine voxels can also be considered as the bounding volumes for collision detection, which is useful in 3D interactions. We also show the potential application of Vox-Surf in scene editing and augmented reality. The source code is publicly available at https://github.com/zju3dv/Vox-Surf.

Achieving precise real-time localization and ensuring robustness are critical challenges in multi-user mobile AR applications. Leveraging collaborative information to augment tracking accuracy on lightweight devices and fortify overall system robustness emerges as a crucial necessity. In this paper, we propose a robust centralized collaborative rnulti-agent VI-SLAM system for mobile AR interaction and server-side efficient consistent mapping. The system deploys a lightweight VIO frontend on mobile devices for real-time tracking, and a backend running on a remote server to update multiple submaps. When overlapping areas between submaps across agents are detected, the system performs submap fusion to establish a globally consistent map. Additionally, we propose a map registration and fusion strategy based on covisibility areas for online registration and fusion in multi-agent scenarios. To improve the tracking accuracy of the frontend on agent, we introduce a strategy for updating the global map to the local map at a moderate frequency between the camera-rate pose estimation of the frontend VIO and the low-frequency global map optimization, using a tightly coupled strategy to achieve consistency of the multi-agent frontend poses estimation in the global map. The effectiveness of the proposed method is further confirmed by executing backend mapping on the server and deploying VIO frontends on multiple mobile devices for AR demostration. Additionally, we discuss the scalability of the proposed system by analyzing network traffic, synchronization frequency, and other factors at both the agent and server ends.

In recent years, the paradigm of neural implicit representations has gained substantial attention in the field of Simultaneous Localization and Mapping (SLAM). However, a notable gap exists in the existing approaches when it comes to scene understanding. In this paper, we introduce NIS-SLAM, an efficient neural implicit semantic RGB-D SLAM system, that leverages a pre-trained 2D segmentation network to learn consistent semantic representations. Specifically, for high-fidelity surface reconstruction and spatial consistent scene understanding, we combine high-frequency multi-resolution tetrahedron-based features and low-frequency positional encoding as the implicit scene representations. Besides, to address the inconsistency of 2D segmentation results from multiple views, we propose a fusion strategy that integrates the semantic probabilities from previous non-keyframes into keyframes to achieve consistent semantic learning. Furthermore, we implement a confidence-based pixel sampling and progressive optimization weight function for robust camera tracking. Extensive experimental results on various datasets show the better or more competitive performance of our system when compared to other existing neural dense implicit RGB-D SLAM approaches. Finally, we also show that our approach can be used in augmented reality applications. Project page: https://zju3dv.github.io/nis_slam.

Visual-inertial SLAM (VI-SLAM) is a key technology for Augmented Reality (AR), which allows the AR device to recover its 6-DoF motion in real-time in order to render the virtual content with the corresponding pose. Nowadays, smartphones are still the mainstream devices for ordinary users to experience AR. However the current VI-SLAM methods, although performing well on high-end phones, still face robustness challenges when deployed on a larger stock of mid- and low-end phones. Existing VI-SLAM datasets use either very ideal sensors or only a limited number of devices for data collection, which cannot reflect the capability gaps that VI-SLAM methods need to solve when deployed on a large variety of phone models. This work proposes 100-Phones. the first VI-SLAM dataset covering a wide range of mainstream phones in the market. The dataset consists of 350 sequences collected by 100 different models of phones. Through analysis and experiments on the collected data, we conclude that the quality of visual-inertial data vary greatly among the mainstream phones, and the current open source VI-SLAM methods still have serious robustness issues when it comes to mass deployment on mobile phones. We release the dataset to facilitate the robustness improvement of VI-SLAM and to promote the mass popularization of AR. Project page: https://github.com/zju3dv/100-Phones.

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.

Local feature matching aims at establishing sparse correspondences between a pair of images. Recently, detector-free methods present generally better performance but are not satisfactory in image pairs with large scale differences. In this paper, we propose Patch Area Transportation with Subdivision (PATS) to tackle this issue. Instead of building an expensive image pyramid, we start by splitting the original image pair into equal-sized patches and gradually resizing and subdividing them into smaller patches with the same scale. However, estimating scale differences between these patches is non-trivial since the scale differences are determined by both relative camera poses and scene structures, and thus spatially varying over image pairs. Moreover, it is hard to obtain the ground truth for real scenes. To this end, we propose patch area transportation, which enables learning scale differences in a self-supervised manner. In contrast to bipartite graph matching, which only handles one-to-one matching, our patch area transportation can deal with many-to-many relationships. PATS improves both matching accuracy and coverage, and shows superior performance in downstream tasks, such as relative pose estimation, visual localization, and optical flow estimation. The source code is available at https://zju3dv.github.io/pats/.

This paper addresses the challenge of capturing performance for the clothed humans from sparse-view or monocular videos. Previous methods capture the performance of full humans with a personalized template or recover the garments from a single frame with static human poses. However, it is inconvenient to extract cloth semantics and capture clothing motion with one-piece template, while single frame-based methods may suffer from instable tracking across videos. To address these problems, we propose a novel method for human performance capture by tracking clothing and human body motion separately with a double-layer neural radiance fields (NeRFs). Specifically, we propose a double-layer NeRFsfor the body and garments, and track the densely deforming template of the clothing and body by jointly optimizing the deformation fields and the canonical double-layer NeRFs. In the optimization, we introduce a physics-aware cloth simulation network which can help generate physically plausible cloth dynamics and body-cloth interactions. Compared with existing methods, our method is fully differentiable and can capture both the body and clothing motion robustly from dynamic videos. Also, our method represents the clothing with an independent NeRFs, allowing us to model implicit fields of general clothes feasibly. The experimental evaluations validate its effectiveness on real multi-view or monocular videos.

Despite the remarkable advances in image matching and pose estimation, image-based localization of a camera in a temporally-varying outdoor environment is still a challenging problem due to huge appearance disparity between query and reference images caused by illumination, seasonal and structural changes. In this work, we propose to leverage additional sensors on a mobile phone, mainly GPS, compass, and gravity sensor, to solve this challenging problem. We show that these mobile sensors provide decent initial poses and effective constraints to reduce the searching space in image matching and final pose estimation. With the initial pose, we are also able to devise a direct 2D-3D matching network to efficiently establish 2D-3D correspondences instead of tedious 2D-2D matching in existing systems. As no public dataset exists for the studied problem, we collect a new dataset that provides a variety of mobile sensor data and significant scene appearance variations, and develop a system to acquire ground-truth poses for query images. We benchmark our method as well as several state-of-the-art baselines and demonstrate the effectiveness of the proposed approach. Our code and dataset are available on the project page: https://zju3dv.github.io/sensloc/

We present PVO, a novel panoptic visual odometry framework to achieve more comprehensive modeling of the scene motion, geometry, and panoptic segmentation information. Our PVO models visual odometry (VO) and video panoptic segmentation (VPS) in a unified view, which makes the two tasks mutually beneficial. Specifically, we introduce a panoptic update module into the VO Module with the guidance of image panoptic segmentation. This Panoptic-Enhanced VO Module can alleviate the impact of dynamic objects in the camera pose estimation with a panoptic-aware dynamic mask. On the other hand, the VO-Enhanced VPS Module also improves the segmentation accuracy by fusing the panoptic segmentation result of the current frame on the fly to the adjacent frames, using geometric information such as camera pose, depth, and optical flow obtained from the VO Module. These two modules contribute to each other through recurrent iterative optimization. Extensive experiments demonstrate that PVO outperforms state-of-the-art methods in both visual odometry and video panoptic segmentation tasks.

Deep hashing-based approaches have become the optimal solutions for large-scale image retrieval task due to their high computational efficiency and low storage burden. Some methods leverage a large teacher network to improve the retrieval performance of the small student network through knowledge distillation, which incurs high computational and time costs. In this paper, we propose Self-Distillation Hashing (SeDH), which improves the image retrieval performance without introducing a complex teacher model and significantly reduces the overall computation costs. Specifically, we generate the soft targets via ensembling the logits of other similar images among the mini-batch. The ensembled soft targets can model the relations between different image samples, which can act as additional supervision for classification. Besides, to learn more compact features and accurate inter-sample similarities, we propose a similarity-preserving loss on the learned hashing features, which aligns the softened similarity distribution with the pairwise soft similarity. Extensive experiments demonstrate that our approach can yield state-of-the-art performance on deep supervised hashing retrieval.

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/.

Stereo depth estimation usually struggles to deal with textureless scenes for both traditional and learning-based methods due to the inherent dependence on image correspondence matching. In this paper, we propose a novel neural network, i.e., DPS-Net, to exploit both the prior geometric knowledge and polarimetric information for depth estimation with two polarimetric stereo images. Specifically, we construct both RGB and polarization correlation volumes to fully leverage the multi-domain similarity between polarimetric stereo images. Since inherent ambiguities exist in the polarization images, we introduce the iso-depth cost explicitly into the network to solve these ambiguities. Moreover, we design a cascaded dual-GRU architecture to recurrently update the disparity and effectively fuse both the multi-domain correlation features and the iso-depth cost. Besides, we present new synthetic and real polarimetric stereo datasets for evaluation. Experimental results demonstrate that our method outperforms the state-of-the-art stereo depth estimation methods.

This paper solves the problem of real-time 6-DoF object tracking from an RGB video. Prior optimization-based methods optimize the object pose by aligning the projected model to the image based on handcrafted features, which are prone to suboptimal solutions. Recent learning-based methods use neural networks to predict the pose, which suffer from limited generalizability or computational efficiency. We propose a learning-based active contour model to make the best use of both worlds. Specifically, given an initial pose, we project the object model to the image plane to obtain the initial contour and use a lightweight network to predict how the contour should move to match the true object boundary, which provides the gradients to optimize the object pose. We also devise an efficient optimization algorithm to train our model end-to-end with pose supervision. Experimental results on semi-synthetic and real-world 6-DoF object tracking datasets demonstrate that our model outperforms state-of-the-art methods by a substantial margin in pose accuracy, while achieving real-time performance on mobile devices. Code is available on our project page: https://zju3dv.github.io/deep_ac/.

Existing inverse rendering combined with neural rendering methods can only perform editable novel view synthesis on object-specific scenes, while we present intrinsic neural radiance fields, dubbed IntrinsicNeRF, which introduce intrinsic decomposition into the NeRF-based neural rendering method and can extend its application to room-scale scenes. Since intrinsic decomposition is a fundamentally under-constrained inverse problem, we propose a novel distance-aware point sampling and adaptive reflectance iterative clustering optimization method, which enables IntrinsicNeRF with traditional intrinsic decomposition constraints to be trained in an unsupervised manner, resulting in multi-view consistent intrinsic decomposition results. To cope with the problem that different adjacent instances of similar reflectance in a scene are incorrectly clustered together, we further propose a hierarchical clustering method with coarse-to-fine optimization to obtain a fast hierarchical indexing representation. It supports compelling real-time augmented applications such as recoloring and illumination variation. Extensive experiments and editing samples on both object-specific/room-scale scenes and synthetic/real-word data demonstrate that we can obtain consistent intrinsic decomposition results and high-fidelity novel view synthesis even for challenging sequences.

Performing accurate localization while maintaining the low-level communication bandwidth is an essential challenge of multi-robot simultaneous localization and mapping (MR-SLAM). In this paper, we tackle this problem by generating a compact yet discriminative feature descriptor with minimum inference time. We propose descriptor distillation that formulates the descriptor generation into a learning problem under the teacher-student framework. To achieve real-time descriptor generation, we design a compact student network and learn it by transferring the knowledge from a pre-trained large teacher model. To reduce the descriptor dimensions from the teacher to the student, we propose a novel loss function that enables the knowledge transfer between two different dimensional descriptors. The experimental results demonstrate that our model is 30% lighter than the state-of-the-art model and produces better descriptors in patch matching. Moreover, we build a MR-SLAM system based on the proposed method and show that our descriptor distillation can achieve higher localization performance for MR-SLAM with lower bandwidth.

Event cameras provide high temporal precision, low data rates, and high dynamic range visual perception, which are well-suited for optical flow estimation. While data-driven optical flow estimation has obtained great success in RGB cameras, its generalization performance is seriously hindered in event cameras mainly due to the limited and biased training data. In this paper, we present a novel simulator, BlinkSim, for the fast generation of large-scale data for event-based optical flow. BlinkSim incorporates a configurable rendering engine alongside an event simulation suite. By leveraging the wealth of current 3D assets, the rendering engine enables us to automatically build up thousands of scenes with different objects, textures, and motion patterns and render very high-frequency images for realistic event data simulation. Based on BlinkSim, we construct a large training dataset and evaluation benchmark BlinkFlow that contains sufficient, diversiform, and challenging event data with optical flow ground truth. Experiments show that BlinkFlow improves the generalization performance of state-of-the-art methods by more than 40% on average and up to 90%. Moreover, we further propose an Event-based optical Flow transFormer (E-FlowFormer) architecture. Powered by our BlinkFlow, E-FlowFormer outperforms the SOTA methods by up to 91% on the MVSEC dataset and 14% on the DSEC dataset and presents the best generalization performance. The source code and data are available at https://zju3dv.github.io/blinkflow/.

In AR applications, the jitter of virtual objects can weaken the sense of integration with the real environment. This jitter is often caused by noise in the pose obtained by 3D tracking or localization methods, especially in monocular vision systems without IMU support. Filtering the pose is an effective method to eliminate jitter, however, it can also cause significant lag in the filtered pose, seriously degrading the AR experience. Existing filters struggle to simultaneously reduce jitter while maintaining low lag. In this paper, we propose a novel Minilag filter, which achieves excellent pose smoothing while significantly reducing the lag through backtracking update and compensation strategies, and has excellent real-time performance. We represent the rotation in the pose in the Lie algebra and filter it in locally Euclidean space, ensuring that the filtering of rotation is consistent with that of vectors. We also analyze the noise distribution and characteristics in the tracked pose, providing a theoretical basis for setting filter parameters. We evaluated the proposed filter using both objective mathematical metrics and a user study, and the experimental results demonstrate that our method achieves state-of-the-art performance.

Recently, Neural Radiance Fields (NeRF) has exhibited significant success in novel view synthesis, surface reconstruction, etc. However, since no physical reflection is considered in its rendering pipeline, NeRF mistakes the reflection in the mirror as a separate virtual scene, leading to the inaccurate reconstruction of the mirror and multi-view inconsistent reflections in the mirror. In this paper, we present a novel neural rendering framework, named Mirror-NeRF, which is able to learn accurate geometry and reflection of the mirror and support various scene manipulation applications with mirrors, such as adding new objects or mirrors into the scene and synthesizing the reflections of these new objects in mirrors, controlling mirror roughness, etc. To achieve this goal, we propose a unified radiance field by introducing the reflection probability and tracing rays following the light transport model of Whitted Ray Tracing, and also develop several techniques to facilitate the learning process. Experiments and comparisons on both synthetic and real datasets demonstrate the superiority of our method. The code and supplementary material are available on the project webpage: https://zju3dv.github.io/Mirror-NeRF/.

This paper presents a collaborative implicit neural simultaneous localization and mapping (SLAM) system with RGB-D image sequences, which consists of complete front-end and back-end modules including odometry, loop detection, sub-map fusion, and global refinement. In order to enable all these modules in a unified framework, we propose a novel neural point based 3D scene representation in which each point maintains a learnable neural feature for scene encoding and is associated with a certain keyframe. Moreover, a distributed-to-centralized learning strategy is proposed for the collaborative implicit SLAM to improve consistency and cooperation. A novel global optimization framework is also proposed to improve the system accuracy like traditional bundle adjustment. Experiments on various datasets demonstrate the superiority of the proposed method in both camera tracking and mapping.

Abstract Visual monocular 6D pose tracking methods for textureless or weakly‐textured objects heavily rely on contour constraints established by the precise 3D model. However, precise models are not always available in reality, and rough models can potentially degrade tracking performance and impede the widespread usage of 3D object tracking. To address this new problem, we propose a novel tracking method that handles rough models. We reshape the rough contour through the probability map, which can avoid explicitly processing the 3D rough model itself. We further emphasize the inner region information of the object, where the points are sampled to provide color constrains. To sufficiently satisfy the assumption of small displacement between frames, the 2D translation of the object is pre‐searched for a better initial pose. Finally, we combine constraints from both the contour and inner region to optimize the object pose. Experimental results demonstrate that the proposed method achieves state‐of‐the‐art performance on both roughly and precisely modeled objects. Particularly for the highly rough model, the accuracy is significantly improved (40.4% v.s. 16.9%).

Abstract We propose RLP‐VIO—a robust and lightweight monocular visual‐inertial odometry system using multiplane priors. With planes extracted from the point cloud, visual‐inertial‐plane PnP uses the plane information for fast localization. Depth estimation is susceptible to degenerated motion, so the planes are expanded in a reprojection consensus‐based way robust to depth errors. For sensor fusion, our sliding‐window optimization uses a novel structureless plane‐distance error cost, which prevents the fill‐in effect that poisons the BA problem's sparsity and permits the use of a smaller sliding window while maintaining good accuracy. The total computational cost is further reduced with our modified marginalization strategy. To further improve the tracking robustness, the landmark depths are constrained using the planes during degenerated motion. The whole system is parallelized with a three‐stage pipeline. Under controlled environments, this parallelization runs deterministically and produces consistent results. The resulting VIO system is tested on widely used datasets and compared with several state‐of‐the‐art systems. Our system achieves competitive accuracy and works robustly even on long and challenging sequences. To demonstrate the effectiveness of the proposed system, we also show the AR application running on mobile devices in real‐time.

Consumer-level RGB-D cameras have been widely used for dense 3D reconstruction of scenes. Especially for textureless or non-lambertian surfaces, consumer RGB-D cameras can ensure completeness of the reconstructed models at a low cost. However, the reconstruction quality relies heavily on the accuracy of the depth sensors. Digital cameras are also used popularly for capturing high-resolution pictures to achieve high-quality dense reconstruction of the scenes, but cannot handle textureless or non-lambertian regions well due to the visual ambiguity problem. To ensure both completeness and accuracy of the reconstructed 3D models, we propose a hybrid multi-view reconstruction pipeline named Hybrid-MVS, which combines the high-resolution images taken by a digital camera and the low-resolution RGB-D frames captured by a consumer RGB-D camera for robust reconstruction of complicated scenes with challenging textureless and non-lambertian surfaces. Unlike most existing multi-sensor systems which require explicit hardware calibration and synchronization of various sensors, the calibration and synchronization problems between the digital camera and RGB-D camera are implicitly solved for compositing reliable depth prior of the digital images in our pipeline. Especially, we propose a hybrid MVS framework for robust PatchMatch stereo and Delaunay meshing, which tightly couples both visual cues given by the digital images and depth cues from the RGB-D frames to maximize the complementary advantages. The experiments with quantitative and qualitative evaluations demonstrate the effectiveness of the proposed Hybrid-MVS framework, which can successfully achieve high-quality 3D reconstruction of complicated natural scenes with robustness to weakly textured and non-lambertian areas.

The combination of augmented reality (AR) and medicine is an important trend in current research. The powerful display and interaction capabilities of the AR system can assist doctors to perform more complex operations. Since the tooth itself is an exposed rigid body structure, dental AR is a relatively hot research direction with application potential. However, none of the existing dental AR solutions are designed for wearable AR devices such as AR glasses. At the same time, these methods rely on high-precision scanning equipment or auxiliary positioning markers, which greatly increases the operational complexity and cost of clinical AR. In this work, we propose a simple and accurate neural-implicit model-driven dental AR system, named ImTooth, and adapted for AR glasses. Based on the modeling capabilities and differentiable optimization properties of state-of-the-art neural implicit representations, our system fuses reconstruction and registration in a single network, greatly simplifying the existing dental AR solutions and enabling reconstruction, registration, and interaction. Specifically, our method learns a scale-preserving voxel-based neural implicit model from multi-view images captured from a textureless plaster model of the tooth. Apart from color and surface, we also learn the consistent edge feature inside our representation. By leveraging the depth and edge information, our system can register the model to real images without additional training. In practice, our system uses a single Microsoft HoloLens 2 as the only sensor and display device. Experiments show that our method can reconstruct high-precision models and accomplish accurate registration. It is also robust to weak, repeating and inconsistent textures. We also show that our system can be easily integrated into dental diagnostic and therapeutic procedures, such as bracket placement guidance.

Accurate localization of a display device is essential for AR in large-scale environments. Visual-based localization is the most commonly used solution, but poses privacy risks, suffers from robustness issues and consumes high power. Wireless signal-based localization is a potential visual-free solution, but its accuracy is not enough for AR. In this paper, we present MagLoc-AR, a novel visual-free localization solution that achieves sufficient accuracy for some AR applications (e.g. AR navigation) in large-scale indoor environments. We exploit the location-dependent magnetic field interference that is ubiquitous indoors as a localization signal. Our method requires only a consumer-grade 9-axis IMU, with the gyroscope and acceleration measurements used to recover the motion trajectory, and the magnetic measurements used to register the trajectory to the global map. To meet the accuracy requirement of AR, we propose a mapping method to reconstruct a globally consistent magnetic field of the environment, and a localization method fusing the biased magnetic measurements with the network-predicted motion to improve localization accuracy. In addition, we provide the first dataset for both visual-based and geomagnetic-based localization in large-scale indoor environments. Evaluations on the dataset demonstrate that our proposed method is sufficiently accurate for AR navigation and has advantages over the visual-based methods in terms of power consumption and robustness. Project page: https://github.com/zju3dv/MagLoc-AR/.

This paper addresses the challenge of human performance capture from sparse multi-view or monocular videos. Given a template mesh of the performer, previous methods capture the human motion by non-rigidly registering the template mesh to images with 2D silhouettes or dense photometric alignment. However, the detailed surface deformation cannot be recovered from the silhouettes, while the photometric alignment suffers from instability caused by appearance variation in the videos. To solve these problems, we propose NerfCap, a novel performance capture method based on the dynamic neural radiance field (NeRF) representation of the performer. Specifically, a canonical NeRF is initialized from the template geometry and registered to the video frames by optimizing the deformation field and the appearance model of the canonical NeRF. To capture both large body motion and detailed surface deformation, NerfCap combines linear blend skinning with embedded graph deformation. In contrast to the mesh-based methods that suffer from fixed topology and texture, NerfCap is able to flexibly capture complex geometry and appearance variation across the videos, and synthesize more photo-realistic images. In addition, NerfCap can be pre-trained end to end in a self-supervised manner by matching the synthesized videos with the input videos. Experimental results on various datasets show that NerfCap outperforms prior works in terms of both surface reconstruction accuracy and novel-view synthesis quality.

This paper addresses the challenge of reconstructing 3D indoor scenes from multi-view images. Many previous works have shown impressive reconstruction results on textured objects, but they still have difficulty in handling low-textured planar regions, which are common in indoor scenes. An approach to solving this issue is to incorporate planer constraints into the depth map estimation in multiview stereo-based methods, but the per-view plane estimation and depth optimization lack both efficiency and multiview consistency. In this work, we show that the planar constraints can be conveniently integrated into the recent implicit neural representation-based reconstruction methods. Specifically, we use an MLP network to represent the signed distance function as the scene geometry. Based on the Manhattan-world assumption, planar constraints are employed to regularize the geometry in floor and wall regions predicted by a 2D semantic segmentation network. To resolve the inaccurate segmentation, we encode the semantics of 3D points with another MLP and design a novel loss that jointly optimizes the scene geometry and semantics in 3D space. Experiments on ScanNet and 7-Scenes datasets show that the proposed method outperforms previous methods by a large margin on 3D reconstruction quality. The code and supplementary materials are available at https://zju3dv.github.io/manhattan_sdf.

We propose a new method named OnePose for object pose estimation. Unlike existing instance-level or category-level methods, OnePose does not rely on CAD models and can handle objects in arbitrary categories without instance-or category-specific network training. OnePose draws the idea from visual localization and only requires a simple RGB video scan of the object to build a sparse SfM model of the object. Then, this model is registered to new query images with a generic feature matching network. To mitigate the slow runtime of existing visual localization methods, we propose a new graph attention network that directly matches 2D interest points in the query image with the 3D points in the SfM model, resulting in efficient and robust pose estimation. Combined with a feature-based pose tracker, OnePose is able to stably detect and track 6D poses of everyday household objects in real-time. We also collected a large-scale dataset that consists of 450 sequences of 150 objects. Code and data are available at the project page: https://zju3dv.github.io/onepose/.

6-DoF object pose estimation from a monocular image is challenging, and a post-refinement procedure is generally needed for high-precision estimation. In this paper, we propose a framework based on a recurrent neural network (RNN) for object pose refinement, which is robust to erroneous initial poses and occlusions. During the recurrent iterations, object pose refinement is formulated as a nonlinear least squares problem based on the estimated correspondence field (between a rendered image and the observed image). The problem is then solved by a differentiable Levenberg-Marquardt (LM) algorithm enabling end-to-end training. The correspondence field estimation and pose refinement are conducted alternatively in each iteration to recover the object poses. Furthermore, to improve the robustness to occlusion, we introduce a consistency-check mechanism based on the learned descriptors of the 3D model and observed 2D images, which downweights the unreliable correspondences during pose optimization. Extensive experiments on LINEMOD, Occlusion-LINEMOD, and YCB-Video datasets validate the effectiveness of our method and demonstrate state-of-the-art performance.

Standard visual localization methods build a priori 3D model of a scene which is used to establish correspondences against the 2D keypoints in a query image. Storing these pre-built 3D scene models can be prohibitively expensive for large-scale environments, especially on mobile devices with limited storage and communication bandwidth. We design a novel framework that compresses a scene while still maintaining localization accuracy. The scene is compressed in three stages: first, the database frames are clustered using pairwise co-visibility information. Then, a learned point selection module prunes the points in each cluster taking into account the final pose estimation accuracy. In the final stage, the features of the selected points are further compressed using learned quantization. Query image registration is done using only the compressed scene points. To the best of our knowledge, we are the first to propose learned scene compression for visual localization. We also demonstrate the effectiveness and efficiency of our method on various outdoor datasets where it can perform accurate localization with low memory consumption.

In this paper, we propose a tightly-coupled SLAM system fused with RGB, Depth, IMU and structured plane information. Traditional sparse points based SLAM systems always maintain a mass of map points to model the environment. Huge number of map points bring us a high computational complexity, making it difficult to be deployed on mobile devices. On the other hand, planes are common structures in man-made environment especially in indoor environments. We usually can use a small number of planes to represent a large scene. So the main purpose of this article is to decrease the high complexity of sparse points based SLAM. We build a lightweight back-end map which consists of a few planes and map points to achieve efficient bundle adjustment (BA) with an equal or better accuracy. We use homography constraints to eliminate the parameters of numerous plane points in the optimization and reduce the complexity of BA. We separate the parameters and measurements in homography and point-to-plane constraints and compress the measurements part to further effectively im-prove the speed of BA. We also integrate the plane information into the whole system to realize robust planar feature extraction, data association, and global consistent planar reconstruction. Finally, we perform an ablation study and compare our method with similar methods in simulation and real environment data. Our system achieves obvious advantages in accuracy and efficiency. Even if the plane parameters are involved in the optimization, we effectively simplify the back-end map by using planar structures. The global bundle adjustment is nearly 2 times faster than the sparse points based SLAM algorithm.

3D environment mapping has been actively stud-ied recently with the development of autonomous driving and augmented reality. Although many image-based methods are proposed due to their convenience and flexibility compared to other complex sensors, few works focus on fixing the inherent scale ambiguity of image-based methods and registering the reconstructed structure to the real-world 3D map, which is very important for autonomous driving. This paper presents a low-cost mapping solution that is able to refine and align the monocular reconstructed point cloud given a public street map. Specifically, we first find the association between the street map and the reconstructed point cloud structure by a novel graph-based geolocalization method. Then, optimized with the corresponding relationship, the map accuracy is significantly improved. The rich environment information can also be associated with the point cloud by the geographical location. Experiments show that our geolocalization algorithm can locate the scene on a gigantic city-scale map (173.46 km2) in two minutes and support 3D map reconstruction with absolute scale and rich environmental information from Internet videos.

We describe a novel approach to indoor place recognition from RGB point clouds based on aggregating low-level colour and geometry features with high-level implicit semantic features. It uses a 2-stage deep learning framework, in which the first stage is trained for the auxiliary task of semantic segmentation and the second stage uses features from layers in the first stage to generate discriminate descriptors for place recognition. The auxiliary task encourages the features to be semantically meaningful, hence aggregating the geometry and colour in the RGB point cloud data with implicit semantic information. We use an indoor place recognition dataset derived from the ScanNet dataset for training and evaluation, with a test set comprising 3,608 point clouds generated from 100 different rooms. Comparison with a traditional feature-based method and four state-of-the-art deep learning methods demonstrate that our approach significantly outperforms all five methods, achieving, for example, a top-3 average recall rate of 75% compared with 41% for the closest rival method. Our code is available at: https://github.com/YuhangMing/Semantic-Indoor-Place-Recognition

In this work, we present a dense tracking and mapping system named Vox-Fusion, which seamlessly fuses neural implicit representations with traditional volumetric fusion methods. Our approach is inspired by the recently developed implicit mapping and positioning system and further extends the idea so that it can be freely applied to practical scenarios. Specifically, we leverage a voxel-based neural implicit surface representation to encode and optimize the scene inside each voxel. Furthermore, we adopt an octree-based structure to divide the scene and support dynamic expansion, enabling our system to track and map arbitrary scenes without knowing the environment like in previous works. Moreover, we proposed a high-performance multi-process framework to speed up the method, thus supporting some applications that require real-time performance. The evaluation results show that our methods can achieve better accuracy and completeness than previous methods. We also show that our Vox-Fusion can be used in augmented reality and virtual reality applications. Our source code is publicly available at https://github.com/zju3dv/Vox-Fusion.

Expanding an existing tourist photo from a partially captured scene to a full scene is one of the desired experiences for photography applications. Although photo extrapolation has been well studied, it is much more challenging to extrapolate a photo (i.e., selfie) from a narrow field of view to a wider one while maintaining a similar visual style. In this paper, we propose a factorized neural re-rendering model to produce photorealistic novel views from cluttered outdoor Internet photo collections, which enables the applications including controllable scene re-rendering, photo extrapolation and even extrapolated 3D photo generation. Specifically, we first develop a novel factorized re-rendering pipeline to handle the ambiguity in the decomposition of geometry, appearance and illumination. We also propose a composited training strategy to tackle the unexpected occlusion in Internet images. Moreover, to enhance photo-realism when extrapolating tourist photographs, we propose a novel realism augmentation process to complement appearance details, which automatically propagates the texture details from a narrow captured photo to the extrapolated neural rendered image. The experiments and photo editing examples on outdoor scenes demonstrate the superior performance of our proposed method in both photo-realism and downstream applications. Code and the supplementary material are available on the project webpage: https://zju3dv.github.io/neural_outdoor_rerender/.

In this paper, we propose a novel system named Disp R-CNN for 3D object detection from stereo images. Many recent works solve this problem by first recovering point clouds with disparity estimation and then apply a 3D detector. The disparity map is computed for the entire image, which is costly and fails to leverage category-specific prior. In contrast, we design an instance disparity estimation network (iDispNet) that predicts disparity only for pixels on objects of interest and learns a category-specific shape prior for more accurate disparity estimation. To address the challenge from scarcity of disparity annotation in training, we propose to use a statistical shape model to generate dense disparity pseudo-ground-truth without the need of LiDAR point clouds, which makes our system more widely applicable. Experiments on the KITTI dataset show that, when LiDAR ground-truth is not used at training time, Disp R-CNN outperforms previous state-of-the-art methods based on stereo input by 20 percent in terms of average precision for all categories. The code and pseudo-ground-truth data are available at the project page: https://github.com/zju3dv/disprcnn.

We tackle the problem of estimating correspondences from a general marker, such as a movie poster, to an image that captures such a marker. Conventionally, this problem is addressed by fitting a homography model based on sparse feature matching. However, they are only able to handle plane-like markers and the sparse features do not sufficiently utilize appearance information. In this paper, we propose a novel framework NeuralMarker, training a neural network estimating dense marker correspondences under various challenging conditions, such as marker deformation, harsh lighting, etc. Deep learning has presented an excellent performance in correspondence learning once provided with sufficient training data. However, annotating pixel-wise dense correspondence for training marker correspondence is too expensive. We observe that the challenges of marker correspondence estimation come from two individual aspects: geometry variation and appearance variation. We, therefore, design two components addressing these two challenges in NeuralMarker. First, we create a synthetic dataset FlyingMarkers containing marker-image pairs with ground truth dense correspondences. By training with FlyingMarkers, the neural network is encouraged to capture various marker motions. Second, we propose the novel Symmetric Epipolar Distance (SED) loss, which enables learning dense correspondence from posed images. Learning with the SED loss and the cross-lighting posed images collected by Structure-from-Motion (SfM), NeuralMarker is remarkably robust in harsh lighting environments and avoids synthetic image bias. Besides, we also propose a novel marker correspondence evaluation method circumstancing annotations on real marker-image pairs and create a new benchmark. We show that NeuralMarker significantly outperforms previous methods and enables new interesting applications, including Augmented Reality (AR) and video editing.

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/.

In this paper, we present a novel monocular visual-inertial odometry system with pre-built maps deployed on the remote server, which can robustly run in real-time on a mobile device even in high latency situations. By tightly coupling VIO with geometric priors from pre-built maps, our system can tolerate the high latency and low frequency of global localization service, which is especially suitable for practical applications when the localization service is deployed on the remote server. Firstly, sparse point clouds are obtained from the dense mesh by the ray casting method according to the localization results. The dense mesh can be reconstructed from the point clouds generated by Structure-from-Motion. We directly use the sparse point clouds in feature tracking and state update to suppress drift. In the process of feature tracking, the high local accuracy of VIO is fully utilized to effectively remove outliers and make our system robust. The experiments on EurocMav datasets and simulation datasets show that compared with state-of-the-art methods, our method can achieve better results in terms of both precision and robustness. The effectiveness of the proposed method is further demonstrated through a real-time AR demo on a mobile phone with the aid of visual localization on the remote server.

Bundle adjustment (BA) is widely used in SLAM and SfM, which are key technologies in Augmented Reality. For real-time SLAM and large-scale SfM, the efficiency of BA is of great importance. This paper proposes CoLi-BA, a novel and efficient BA solver that significantly improves the optimization speed by compact linearization and reordering. Specifically, for each reprojection function, the redundant matrix representation of Jacobian is replaced with a tiny 3D vector, by which the computational complexity, memory storage, and cache missing for Hessian matrix construction and Schur complement are significantly reduced. Besides, we also propose a novel reordering strategy to improve the cache efficiency for Schur complement. Experiments on diverse datasets show that the speed of the proposed CoLi-BA is five times that of Ceres and two times that of g2o without sacrificing accuracy. We further verify the effectiveness by porting CoLi-BA to the open-source SLAM and SfM systems. Even when running the proposed solver in a single thread, the local BA of SLAM only takes about 20ms on a desktop PC, and the reconstruction of SfM with seven thousand photos only takes half an hour. The source code is available on the webpage: https://github.com/zju3dv/CoLi-BA.

3D representation and reconstruction of human bodies have been studied for a long time in computer vision. Traditional methods rely mostly on parametric statistical linear models, limiting the space of possible bodies to linear combinations. It is only recently that some approaches try to leverage neural implicit representations for human body modeling, and while demonstrating impressive results, they are either limited by representation capability or not physically meaningful and controllable. In this work, we propose a novel neural implicit representation for the human body, which is fully differentiable and optimizable with disentangled shape and pose latent spaces. Contrary to prior work, our representation is designed based on the kinematic model, which makes the representation controllable for tasks like pose animation, while simultaneously allowing the optimization of shape and pose for tasks like 3D fitting and pose tracking. Our model can be trained and fine-tuned directly on non-watertight raw data with well-designed losses. Experiments demonstrate the improved 3D reconstruction performance over SoTA approaches and show the applicability of our method to shape interpolation, model fitting, pose tracking, and motion retargeting.

Visual localization is of great importance in robotics and computer vision. Recently, scene coordinate regression based methods have shown good performance in visual localization in small static scenes. However, it still estimates camera poses from many inferior scene coordinates. To address this problem, we propose a novel visual localization framework that establishes 2D-to-3D correspondences between the query image and the 3D map with a series of learnable scene-specific landmarks. In the landmark generation stage, the 3D surfaces of the target scene are over-segmented into mosaic patches whose centers are regarded as the scene-specific landmarks. To robustly and accurately recover the scene-specific landmarks, we propose the Voting with Segmentation Network (VS-Net) to segment the pixels into different landmark patches with a segmentation branch and estimate the landmark locations within each patch with a landmark location voting branch. Since the number of landmarks in a scene may reach up to 5000, training a segmentation network with such a large number of classes is both computation and memory costly for the commonly used cross-entropy loss. We propose a novel prototype-based triplet loss with hard negative mining, which is able to train semantic segmentation networks with a large number of labels efficiently. Our proposed VS-Net is extensively tested on multiple public benchmarks and can outperform stateof-the-art visual localization methods. Code and models are available at https://github.com/zju3dv/VS-Net.

Different from traditional video cameras, event cam- eras capture asynchronous events stream in which each event encodes pixel location, trigger time, and the polarity of the brightness changes. In this paper, we introduce a novel graph-based framework for event cameras, namely SlideGCN. Unlike some recent graph-based methods that use groups of events as input, our approach can efficiently process data event-by-event, unlock the low latency nature of events data while still maintaining the graph’s structure internally. For fast graph construction, we develop a radius search algorithm, which better exploits the partial regular structure of event cloud against k-d tree based generic methods. Experiments show that our method reduces the computational complexity up to 100 times with respect to current graph-based methods while keeping state-of-the-art performance on object recognition. Moreover, we verify the superiority of event-wise processing with our method. When the state becomes stable, we can give a prediction with high confidence, thus making an early recognition.

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.

Real-time dense reconstruction has been extensively studied for its wide applications in computer vision and robotics, meanwhile much effort has been made for the multi-robot system which plays an irreplaceable role in complicated but time-critical scenarios, e.g., search and rescue tasks. In this paper, we propose an efficient system named Coxgraph for multi-robot collaborative dense reconstruction in real-time. In our system, each client performs volumetric mapping in a producer-consumer manner. To facilitate transmission, we propose a compact 3D representation which transforms the SDF submap to mesh packs. During the recovery of submaps from mesh packs, the system can perform loop closure outlier rejection based on geometry consistency, trajectory collision and fitness check. Then we develop a robust map fusion method through joint optimization of trajectories and submaps. Extensive experiments demonstrate that our system can produce a globally consistent dense map in real-time with less transmission load, which is available as open-source software1.

In this work, we propose a tightly-coupled EKF framework for visual-inertial odometry with NIN (Neural Inertial Navigation) aided. Traditional VIO systems are fragile in challenging scenes with weak or confusing visual information, such as weak/repeated texture, dynamic environment, fast camera motion with serious motion blur, etc. It is extremely difficult for a vision-based algorithm to handle these problems. So we firstly design a robust deep learning based inertial network (called RNIN), using only IMU measurements as input. RNIN is significantly more robust in challenging scenes than traditional VIO systems. In order to take full advantage of vision-based algorithms in AR/VR areas, we further develop a multi-sensor fusion system RNIN-VIO, which tightly couples the visual, IMU and NIN measurements. Our system performs robustly in extremely challenging conditions, with high precision both in trajectories and AR effects. The experimental results of evaluation on dataset evaluation and online AR demo demonstrate the superiority of the proposed system in robustness and accuracy.

Robust and accurate global 6DoF localization is essential for many applications, i.e., augmented reality and autonomous driving. Most existing 6DoF visual localization approaches need to build a dense texture model in advance, which is computationally extensive and almost infeasible in the global range. In this work, we propose BDLoc, a hierarchical global localization framework via the 2.5D building map, which is able to estimate the accurate pose of the query street-view image without using detailed dense 3D model and texture information. Specifically speaking, we first extract the 3D building information from the street-view image and surrounding 2.5D building map, and then solve a coarse relative pose by local to global registration. In order to improve the feature extraction, we propose a novel SPG-Net which is able to capture both local and global features. Finally, an iterative semantic alignment is applied to obtain a finner result with the differentiable rendering and the cross-view semantic constraint. Except for a coarse longitude and latitude from GPS, BDLoc doesn’t need any additional information like altitude and orientation that are necessary for many previous works. We also create a large dataset to explore the performance of the 2.5D map-based localization task. Extensive experiments demonstrate the superior performance of our method.

This paper presents a novel framework to estimate parametric model- s for 3D clothed humans from partial point clouds. It is a challenging problem due to factors such as arbitrary human shape and pose, large variations in clothing details, and significant missing data. Existing methods mainly focus on estimating the parametric model of undressed bodies or reconstructing the non-parametric 3D shapes from point clouds. In this paper, we propose a hierarchical regression framework to learn the parametric model of detailed human shapes from partial point clouds of a single depth frame. Benefiting from the favorable ability of deep neural networks to model nonlinearity, the proposed framework cascades several successive regression networks to estimate the parameters of detailed 3D human body models in a coarse-to-fine manner. Specifically, the first global regression network extracts global deep features of point clouds to obtain an initial estimation of the undressed human model. Based on the initial estimation, the local regression network then refines the undressed human model by using the local features of neighborhood points of human joints. Finally, the clothing details are inferred as an additive displacement on the refined undressed model using the vertex-level regression network. The experimental results demonstrate that the proposed hierarchical regression approach can accurately predict detailed human shapes from partial point clouds and outperform prior works in the recovery accuracy of 3D human models.

We present a novel end-to-end plane detection and description network named SuperPlane to detect and match planes in two RGB images. SuperPlane takes a single image as input and extracts 3D planes and generates corresponding descriptors simultaneously. A mask-attention module and an instance-triplet loss are proposed to improve the distinctiveness of the plane descriptor. For image matching, we also propose an area-aware Kullback-Leibler (KL) divergence retrieval method. Extensive experiments show that the proposed method outperforms state-of-the-art methods and retains good generalization capacity. The applications in image-based localization and augmented reality also demonstrate the effectiveness of SuperPlane.

We propose a novel deep learning method to predict dense correspondences for partial point clouds of non-rigidly deformable targets. Dense correspondences are learned in the form of vertex displacements of a template mesh towards the point clouds. A two-stage regression framework is proposed to estimate accurate displacement vectors, including the global and local regression networks. Specifically, the global regression network estimates global displacements from the global features of the template mesh and point clouds through a graph CNN based hierarchical encoder-decoder network. Based on the initial displacements, a mesh can be generated that fits to the point clouds roughly. In the local regression network, a local feature embedding layer fuses local features of point clouds with graph features on the generated mesh through an attention mechanism. Consequently, the embedded local features are employed to refine the correspondences in local regions of the targets by predicting the increments of vertex displacements. Our method is further generalized to correspondence estimation on unseen real data with a robust fine-tuning method. The experimental results on diverse datasets of various deformable subjects (e.g., human bodies, animals, and hands) demonstrate that the proposed approach can accurately and robustly estimate dense correspondences from non-rigid point clouds.

We present a novel online 3D scanning system for high-quality object reconstruction with a mobile device, called Mobile3DScanner. Using a mobile device equipped with an embedded RGBD camera, our system provides online 3D object reconstruction capability for users to acquire high-quality textured 3D object models. Starting with a simultaneous pose tracking and TSDF fusion module, our system allows users to scan an object with a mobile device to get a 3D model for real-time preview. After the real-time scanning process is completed, the scanned 3D model is globally optimized and mapped with multi-view textures as an efficient postprocess to get the final textured 3D model on the mobile device. Unlike most existing state-of-the-art systems which can only scan homeware objects such as toys with small dimensions due to the limited computation and memory resources of mobile platforms, our system can reconstruct objects with large dimensions such as statues. We propose a novel visual-inertial ICP approach to achieve real-time accurate 6DoF pose tracking of each incoming frame on the front end, while maintaining a keyframe pool on the back end where the keyframe poses are optimized by local BA. Simultaneously, the keyframe depth maps are fused by the optimized poses to a TSDF model in real-time. Especially, we propose a novel adaptive voxel resizing strategy to solve the out-of-memory problem of large dimension TSDF fusion on mobile platforms. In the post-process, the keyframe poses are globally optimized and the keyframe depth maps are optimized and fused to obtain a final object model with more accurate geometry. The experiments with quantitative and qualitative evaluation demonstrate the effectiveness of the proposed 3D scanning system based on a mobile device, which can successfully achieve online high-quality 3D reconstruction of natural objects with larger dimensions for efficient AR content creation.

In this paper, we present a novel deep architecture to recover a 3D shape in triangular mesh from a single image based on mesh deformation. Most existing deformation-based methods produce uniform mesh predictions by repeatedly applying global subdivision but fail to require the highlighted details due to the memory limits. To address this problem, we propose a novel saliency guided subdivision method to achieve the trade-off between detail generation and memory consumption. Instead of using local geometric cues such as curvature, we introduce a global point-based saliency voting operation to guide the adaptive mesh subdivision and deformation explicitly. Moreover, we propose the oriented chamfer loss to mitigate the mesh self-intersection problem in subdivision. We further make our network configurable and explore the best structure combination. Extensive experiments show that our method can both produce visually pleasing results with fine details and achieve better performance compared to other state-of-the-art methods.

This work addresses the problem of 3D human pose and shape estimation from a sequence of point clouds. Existing sequential 3D human shape estimation methods mainly focus on the template model fitting from a sequence of depth images or the parametric model regression from a sequence of RGB images. In this paper, we propose a novel sequential 3D human pose and shape estimation framework from a sequence of point clouds. Specifically, the proposed framework can regress 3D coordinates of mesh vertices at different resolutions from the latent features of point clouds. Based on the estimated 3D coordinates and features at the low resolution, we develop a spatial-temporal mesh attention convolution (MAC) to predict the 3D coordinates of mesh vertices at the high resolution. By assigning specific attentional weights to different neighboring points in the spatial and temporal domains, our spatial-temporal MAC can capture structured spatial and temporal features of point clouds. We further generalize our framework to the real data of human bodies with a weakly supervised fine-tuning method. The experimental results on SURREAL, Human3.6M, DFAUST and the real detailed data demonstrate that the proposed approach can accurately recover the 3D body model sequence from a sequence of point clouds.

Obtaining accurate and sufficient feature matches is crucial for robust large-scale Structure-from-Motion. For unordered image collections, a traditional feature matching method with geometric verification requires a huge cost to find sufficient feature matches. Although several methods have been proposed to speed up this stage, none of them makes full use of existing matches. In this paper, we propose a novel efficient image matching method by using the transitivity of region covisibility. The overlapping image pairs can be efficiently found in an iterative matching strategy even only with few inlier feauture matches. The experimental results on unordered image datasets demonstrate that the proposed method is three times faster than the state-of-the-art and the matching result is high-quality enough for robust SfM.

2D-3D matching is an essential step for visual localization, where the accuracy of the camera pose is mainly determined by the quality of 2D-3D correspondences. The matching is typically achieved by the nearest neighbor search of local features. Many existing works have shown impressive results on both the efficiency and accuracy. Recently emerged learning-based features further improve the robustness compared to the traditional hand-crafted ones. However, it is still hard to establish enough correct matches in challenging scenes with illumination changes or repetitive patterns due to the intrinsic local properties of local features. In this work, we propose a novel method to deal with 2D-3D matching in a very robust way. We first establish as many potential correct matches as possible using the local similarity. Then we construct a bipartite graph and use a deep neural network, referred to as Bipartite Graph Network (BGNet), to extract the global geometric information. The network predicts the likelihood of being an inlier for each edge and outputs the globally optimal one-to-one correspondences with a Hungarian pooling layer. The experiments show that the proposed method can find more correct matches and improves localization on both the robustness and accuracy. The results on multiple visual localization datasets are obviously better than the existing state-of-the-arts, which demonstrates the effectiveness of the proposed method.

Intrinsic image decomposition, i.e., decomposing a natural image into a reflectance image and a shading image, is used in many augmented reality applications for achieving better visual coherence between virtual contents and real scenes. The main challenge is that the decomposition is ill-posed, especially in indoor scenes where lighting conditions are complicated, while real training data is inadequate. To solve this challenge, we propose NIID-Net, a novel learning-based framework that adapts surface normal knowledge for improving the decomposition. The knowledge learned from relatively more abundant data for surface normal estimation is integrated into intrinsic image decomposition in two novel ways. First, normal feature adapters are proposed to incorporate scene geometry features when decomposing the image. Secondly, a map of integrated lighting is proposed for propagating object contour and planarity information during shading rendering. Furthermore, this map is capable of representing spatially-varying lighting conditions indoors. Experiments show that NIID-Net achieves competitive performance in reflectance estimation and outperforms all previous methods in shading estimation quantitatively and qualitatively. The source code of our implementation is released at https://github.com/zju3dv/NIID-Net.

We present a real-time monocular 3D reconstruction system on a mobile phone, called Mobile3DRecon. Using an embedded monocular camera, our system provides an online mesh generation capability on back end together with real-time 6DoF pose tracking on front end for users to achieve realistic AR effects and interactions on mobile phones. Unlike most existing state-of-the-art systems which produce only point cloud based 3D models online or surface mesh offline, we propose a novel online incremental mesh generation approach to achieve fast online dense surface mesh reconstruction to satisfy the demand of real-time AR applications. For each keyframe of 6DoF tracking, we perform a robust monocular depth estimation, with a multi-view semi-global matching method followed by a depth refinement post-processing. The proposed mesh generation module incrementally fuses each estimated keyframe depth map to an online dense surface mesh, which is useful for achieving realistic AR effects such as occlusions and collisions. We verify our real-time reconstruction results on two mid-range mobile platforms. The experiments with quantitative and qualitative evaluation demonstrate the effectiveness of the proposed monocular 3D reconstruction system, which can handle the occlusions and collisions between virtual objects and real scenes to achieve realistic AR effects.

Camera localization from monocular images has been a long-standing problem, but its robustness in dynamic environments is still not adequately addressed. Compared with classic geometric approaches, modern CNN-based methods (e.g. PoseNet) have manifested the reliability against illumination or viewpoint variations, but they still have the following limitations. First, foreground moving objects are not explicitly handled, which results in poor performance and instability in dynamic environments. Second, the output for each image is a point estimate without uncertainty quantification. In this paper, we propose a framework which can be generally applied to existing CNN-based pose regressors to improve their robustness in dynamic environments. The key idea is a prior guided dropout module coupled with a self-attention module which can guide CNNs to ignore foreground objects during both training and inference. Additionally, the dropout module enables the pose regressor to output multiple hypotheses from which the uncertainty of pose estimates can be quantified and leveraged in the following uncertainty-aware pose graph optimization to improve the robustness further. We achieve an average accuracy of 9.98m/3.63° on RobotCar dataset, which outperforms the state-of-the-art method by 62.97%/47.08%. The source code of our implementation is available at https://github.com/zju3dv/RVL-Dynamic.

Depth completion aims to recover dense depth maps from sparse depth measurements. It is of increasing importance for autonomous driving and draws increasing attention from the vision community. Most of the current competitive methods directly train a network to learn a mapping from sparse depth inputs to dense depth maps, which has difficulties in utilizing the 3D geometric constraints and handling the practical sensor noises. In this paper, to regularize the depth completion and improve the robustness against noise, we propose a unified CNN framework that 1) models the geometric constraints between depth and surface normal in a diffusion module and 2) predicts the confidence of sparse LiDAR measurements to mitigate the impact of noise. Specifically, our encoder-decoder backbone predicts the surface normal, coarse depth and confidence of LiDAR inputs simultaneously, which are subsequently inputted into our diffusion refinement module to obtain the final completion results. Extensive experiments on KITTI depth completion dataset and NYU-Depth-V2 dataset demonstrate that our method achieves state-of-the-art performance. Further ablation study and analysis give more insights into the proposed components and demonstrate the generalization capability and stability of our model.

Monocular visual-inertial tracking without good initialization easily fails due to its non-linear nature. Rapid and accurate metric initialization is crucial. In this paper, we propose a novel monocular visual-inertial initialization method which can initialize the IMU states, camera poses, and scale in a rapid and robust way. To avoid mixing gravity and accelerometer bias, we propose to use the detected vertical edges to estimate a better gravity. This improves the observability to the underlying problem even without sufficient movement, so we can solve all the states crucial for a good initialization. We evaluate our approach on EuRoC dataset and compare with existing state-of-the-art methods. The experimental results demonstrate the effectiveness of the proposed method.

Although VSLAM/VISLAM has achieved great success, it is still difficult to quantitatively evaluate the localization results of different kinds of SLAM systems from the aspect of augmented reality due to the lack of an appropriate benchmark. For AR applications in practice, a variety of challenging situations (e.g., fast motion, strong rotation, serious motion blur, dynamic interference) may be easily encountered since a home user may not carefully move the AR device, and the real environment may be quite complex. In addition, the frequency of camera lost should be minimized and the recovery from the failure status should be fast and accurate for good AR experience. Existing SLAM datasets/benchmarks generally only provide the evaluation of pose accuracy and their camera motions are somehow simple and do not fit well the common cases in the mobile AR applications. With the above motivation, we build a new visual-inertial dataset as well as a series of evaluation criteria for AR. We also review the existing monocular VSLAM/VISLAM approaches with detailed analyses and comparisons. Especially, we select 8 representative monocular VSLAM/VISLAM approaches/systems and quantitatively evaluate them on our benchmark. Our dataset, sample code and corresponding evaluation tools are available at the benchmark website http://www.zjucvg.net/eval-vislam/.

Modern visual-inertial SLAM (VI-SLAM) achieves higher accuracy and robustness than pure visual SLAM, thanks to the complementariness of visual features and inertial measurements. However, jointly using visual and inertial measurements to optimize SLAM objective functions is a problem of high computational complexity. In many VI-SLAM applications, the conventional optimization solvers can only use a very limited number of recent measurements for real time pose estimation, at the cost of suboptimal localization accuracy. In this work, we renovate the numerical solver for VI-SLAM. Compared to conventional solvers, our proposal provides an exact solution with significantly higher computational efficiency. Our solver allows us to use remarkably larger number of measurements to achieve higher accuracy and robustness. Furthermore, our method resolves the global consistency problem that is unaddressed by many state-of-the-art SLAM systems: to guarantee the minimization of re-projection function and inertial constraint function during loop closure. Experiments demonstrate our novel formulation renders lower localization error and more than 10x speedup compared to alternatives. We release the source code of our implementation to benefit the community1.

In this work, we develop a novel keyframe-based dense planar SLAM (KDP-SLAM) system, based on CPU only, to reconstruct large indoor environments in real-time using a hand-held RGB-D sensor. Our keyframe-based approach applies a fast dense method to estimate odometry, fuses depth measurements from small baseline images, extracts planes from the fused depth map, and optimizes the poses of the keyframes and landmark planes in a global factor graph using incremental smoothing and mapping (iSAM). Using the fast odometry estimation, correct plane correspondences may be found projectively, and the pose of each frame can be estimated accurately even without sufficient planes to fully constrain the 6 degree-of-freedom transformation. The depth map generated from the local fusion process generates higher quality reconstructions and plane segmentations by eliminating noise. Moreover, explicitly modeling plane landmarks in the fully probabilistic global optimization significantly reduces the drift that plagues other dense SLAM algorithms. We test our system on standard RGB-D benchmarks as well as additional indoor environments, demonstrating its state-of-the-art performance as a real-time dense 3D SLAM algorithm, without the use of GPU.

Traditional stereo matching approaches generally have problems in handling textureless regions, strong occlusions and reflective regions that do not satisfy a Lambertian surface assumption. In this paper, we propose to combine the predicted surface normal by deep learning to overcome these inherent difficulties in stereo matching. With the selected reliable disparities from stereo matching method and effective edge fusion strategy, we can faithfully convert the predicted surface normal map to a disparity map by solving a least squares system which maintains discontinuity on object boundaries and continuity on other regions. Then we refine the disparity map iteratively by bilateral filtering-based completion and edge feature refinement. Experimental results on the Middlebury dataset and our own captured stereo sequences demonstrate the effectiveness of the proposed approach.

We present a novel templateless approach for nonrigid reconstruction and motion tracking using a single RGB-D camera. Without any template prior, our system achieves accurate reconstruction and tracking for considerably deformable objects. To robustly register the input sequence of partial depth scans with dynamic motion, we propose an efficient local-to-global hierarchical optimization framework inspired by the idea of traditional structure-from-motion. Our proposed framework mainly consists of two stages, local nonrigid bundle adjustment and global optimization. To eliminate error accumulation during the nonrigid registration of loop motion sequences, we split the full sequence into several segments and apply local nonrigid bundle adjustment to align each segment locally. Global optimization is then adopted to combine all segments and handle the drift problem through loop-closure constraint. By fitting to the input partial data, a deforming 3D model sequence of dynamic objects is finally generated. Experiments on both synthetic and real test data sets and comparisons with state of the art demonstrate that our approach can handle considerable motions robustly and efficiently, and reconstruct high-quality 3D model sequences without drift.

Keyframe-based SLAM has achieved great success in terms of accuracy, efficiency and scalability. However, due to parallax requirement and delay of map expansion, traditional keyframe-based methods easily encounter the robustness problem in the challenging cases especially for fast motion with strong rotation. For AR applications in practice, these challenging cases are easily encountered, since a home user may not carefully move the camera to avoid potential problems. With the above motivation, in this paper, we present RKSLAM, a robust keyframe-based monocular SLAM system that can reliably handle fast motion and strong rotation, ensuring good AR experiences. First, we propose a novel multihomography based feature tracking method which is robust and efficient for fast motion and strong rotation. Based on it, we propose a real-time local map expansion scheme to triangulate the observed 3D points immediately without delay. A sliding-window based camera pose optimization framework is proposed, which imposes the motion prior constraints between consecutive frames through simulated or real IMU data. Qualitative and quantitative comparisons with the state-of-the-art methods, and an AR application on mobile devices demonstrate the effectiveness of the proposed approach.

We present a novel image stitching approach, which can produce visually plausible panoramic images with input taken from different viewpoints. Unlike previous methods, our approach allows wide baselines between images and non-planar scene structures. Instead of 3D reconstruction, we design a mesh-based framework to optimize alignment and regularity in 2D. By solving a global objective function consisting of alignment and a set of prior constraints, we construct panoramic images, which are locally as perspective as possible and yet nearly orthogonal in the global view. We improve composition and achieve good performance on misaligned areas. Experimental results on challenging data demonstrate the effectiveness of the proposed method.

Structure-from-motion (SfM) largely relies on feature tracking. In image sequences, if disjointed tracks caused by objects moving in and out of the field of view, occasional occlusion, or image noise are not handled well, corresponding SfM could be affected. This problem becomes severer for large-scale scenes, which typically requires to capture multiple sequences to cover the whole scene. In this paper, we propose an efficient non-consecutive feature tracking framework to match interrupted tracks distributed in different subsequences or even in different videos. Our framework consists of steps of solving the feature "dropout" problem when indistinctive structures, noise or large image distortion exists, and of rapidly recognizing and joining common features located in different subsequences. In addition, we contribute an effective segment-based coarse-to-fine SfM algorithm for robustly handling large data sets. Experimental results on challenging video data demonstrate the effectiveness of the proposed system.

Extracting spatio-temporally consistent segments from a video sequence is a challenging problem due to the complexity of color, motion and occlusions. Most existing spatio-temporal segmentation approaches have inherent difficulties in handling large displacement with significant occlusions . This paper presents a novel framework for spatio-temporal segmentation. With the estimated depth data beforehand by a multi-view stereo technique, we project the pixels to other frames for collecting the boundary and segmentation statistics in a video, and incorporate them into the segmentation energy for spatio-temporal optimization. In order to effectively solve this problem, we introduce an iterative optimization scheme by first initializing segmentation maps for each frame independently, and then link the correspondences among different frames and iteratively refine them with the collected statistics, so that a set of spatio-temporally consistent volume segments are finally achieved. The effectiveness and usefulness of our automatic framework are demonstrated via its applications for 3D reconstruction, video editing and semantic segmentation on a variety of challenging video examples.

Although multi-view stereo has been extensively studied during the past decades, automatically computing high-quality dense depth information from captured images/videos is still quite difficult. Many factors, such as serious occlusion, large texture less regions and strong reflection, easily cause erroneous depth recovery. In this paper, we present a novel semi-automatic multi-view stereo system, which can quickly create and repair depth from a monocular sequence taken by a freely moving camera. One of our main contributions is that we propose a novel multi-view stereo model incorporating prior constraints indicated by user interaction, which makes it possible to even handle Non-Lambertian surface that surely violates the photo-consistency constraint. Users only need to provide a coarse segmentation and a few user interactions, our system can automatically correct depth and refine boundary. With other priors and occlusion handling, the erroneous depth can be effectively corrected even for very challenging examples that are difficult for state-of-the-art methods.

We present a novel 3D reconstruction approach using a low-cost RGB-D camera such as Microsoft Kinect. Compared with previous methods, our scanning system can work well in challenging cases where there are large repeated textures and significant depth missing problems. For robust registration, we propose to utilize both visual and geometry features and combine SFM technique to enhance the robustness of feature matching and camera pose estimation. In addition, a novel prior-based multicandidates RANSAC is introduced to efficiently estimate the model parameters and significantly speed up the camera pose estimation under multiple correspondence candidates. Even when serious depth missing occurs, our method still can successfully register all frames together. Loop closure also can be robustly detected and handled to eliminate the drift problem. The missing geometry can be completed by combining multiview stereo and mesh deformation techniques. A variety of challenging examples demonstrate the effectiveness of the proposed approach.

We present a novel real-time monocular SLAM system which can robustly work in dynamic environments. Different to the traditional methods, our system allows parts of the scene to be dynamic or the whole scene to gradually change. The key contribution is that we propose a novel online keyframe representation and updating method to adaptively model the dynamic environments, where the appearance or structure changes can be effectively detected and handled. We reliably detect the changed features by projecting them from the keyframes to current frame for appearance and structure comparison. The appearance change due to occlusions also can be reliably detected and handled. The keyframes with large changed areas will be replaced by newly selected frames. In addition, we propose a novel prior-based adaptive RANSAC algorithm (PARSAC) to efficiently remove outliers even when the inlier ratio is rather low, so that the camera pose can be reliably estimated even in very challenging situations. Experimental results demonstrate that the proposed system can robustly work in dynamic environments and outperforms the state-of-the-art SLAM systems (e.g. PTAM).

In this paper, we propose a novel dense depth recovery method for a trinocular video sequence. Specifically, we contribute a novel trinocular stereo matching model, which can effectively utilize the advantages of trinocular stereo images, and incorporate the visibility term with segmentation prior for robust depth estimate. In order to make the recovered depth maps more accurate and temporally consistent, we propose to first classify the pixels to static and dynamic ones, and then perform spatio-temporal depth optimization for them in different ways. Especially, we propose two motion models for handling dynamic pixels. The traditional bundle optimization model and our spatio-temporal optimization model are softly combined in a probabilistic way, so that the depths of both static and dynamic pixels can be effectively refined. Our automatic depth recovery approach is evaluated using a variety of challenging trinocular video sequences.

This paper presents a novel multi-body multi-view stereo method to simultaneously recover dense depth maps and perform segmentation with the input of a monocular image sequence. Unlike traditional multi-view stereo approaches that generally handle a single static scene or an object, we show that depth estimation and segmentation can be jointly modeled and be globally solved in an energy minimization framework for ubiquitous scenes containing multiple independently moving rigid objects. Our major contribution includes a new multi-body stereo model, which integrates the color, geometry, and layer constraints for spatio-temporal depth recovery and automatic object segmentation. A two-pass optimization scheme is proposed to progressively update the estimates. Our method is applied to a variety of challenging examples.

Extracting high-quality dynamic foreground layers from a video sequence is a challenging problem due to the coupling of color, motion, and occlusion. Many approaches assume that the background scene is static or undergoes the planar perspective transformation. In this paper, we relax these restrictions and present a comprehensive system for accurately computing object motion, layer, and depth information. A novel algorithm that combines different clues to extract the foreground layer is proposed, where a voting-like scheme robust to outliers is employed in optimization. The system is capable of handling difficult examples in which the background is nonplanar and the camera freely moves during video capturing. Our work finds several applications, such as high-quality view interpolation and video editing.

In this paper, we present a novel method to extract motion of a dynamic object from a video that is captured by a handheld camera, and apply it to a 3D character. Unlike the motion capture techniques, neither special sensors/trackers nor a controllable environment is required. Our system significantly automates motion imitation which is traditionally conducted by professional animators via manual keyframing. Given the input video sequence, we track the dynamic reference object to obtain trajectories of both 2D and 3D tracking points. With them as constraints, we then transfer the motion to the target 3D character by solving an optimization problem to maintain the motion gradients. We also provide a user-friendly editing environment for users to fine tune the motion details. As casual videos can be used, our system, therefore, greatly increases the supply source of motion data. Examples of imitating various types of animal motion are shown.

We present a novel keyframe selection and recognition method for robust markerless real-time camera tracking. Our system contains an offline module to select features from a group of reference images and an online module to match them to the input live video in order to quickly estimate the camera pose. The main contribution lies in constructing an optimal set of keyframes from the input reference images, which are required to approximately cover the entire space and at the same time minimize the content redundancy amongst the selected frames. This strategy not only greatly saves the computation, but also helps significantly reduce the number of repeated features so as to improve the camera tracking quality. Our system also employs a parallel-computing scheme with multi-CPU hardware architecture. Experimental results show that our method dramatically enhances the computation efficiency and eliminates the jittering artifacts.

Abstract High‐quality video editing usually requires accurate layer separation in order to resolve occlusions. However, most of the existing bilayer segmentation algorithms require either considerable user intervention or a simple stationary camera configuration with known background, which is difficult to meet for many real world online applications. This paper demonstrates that various visually appealing montage effects can be online created from a live video captured by a rotating camera, by accurately retrieving the camera state and segmenting out the dynamic foreground. The key contribution is that a novel fast bilayer segmentation method is proposed which can effectively extract the dynamic foreground under rotational camera configuration, and is robust to imperfect background estimation and complex background colors. Our system can create a variety of live visual effects, including but not limited to, realistic virtual object insertion, background substitution and blurring, non‐photorealistic rendering and camouflage effect. A variety of challenging examples demonstrate the effectiveness of our method.

This paper presents a novel method for recovering consistent depth maps from a video sequence. We propose a bundle optimization framework to address the major difficulties in stereo reconstruction, such as dealing with image noise, occlusions, and outliers. Different from the typical multi-view stereo methods, our approach not only imposes the photo-consistency constraint, but also explicitly associates the geometric coherence with multiple frames in a statistical way. It thus can naturally maintain the temporal coherence of the recovered dense depth maps without over-smoothing. To make the inference tractable, we introduce an iterative optimization scheme by first initializing the disparity maps using a segmentation prior and then refining the disparities by means of bundle optimization. Instead of defining the visibility parameters, our method implicitly models the reconstruction noise as well as the probabilistic visibility. After bundle optimization, we introduce an efficient space-time fusion algorithm to further reduce the reconstruction noise. Our automatic depth recovery is evaluated using a variety of challenging video examples.

Compared to still image editing, content-based video editing faces the additional challenges of maintaining the spatiotemporal consistency with respect to geometry. This brings up difficulties of seamlessly modifying video content, for instance, inserting or removing an object. In this paper, we present a new video editing system for creating spatiotemporally consistent and visually appealing refilming effects. Unlike the typical filming practice, our system requires no labor-intensive construction of 3D models/surfaces mimicking the real scene. Instead, it is based on an unsupervised inference of view-dependent depth maps for all video frames. We provide interactive tools requiring only a small amount of user input to perform elementary video content editing, such as separating video layers, completing background scene, and extracting moving objects. These tools can be utilized to produce a variety of visual effects in our system, including but not limited to video composition, "predator" effect, bullet-time, depth-of-field, and fog synthesis. Some of the effects can be achieved in real time.

This paper presents a novel method for reconstructing high-quality video depth maps. A bundle optimization model is proposed to address the key issues, including image noise and occlusions, in stereo reconstruction. Our method not only uses the color constancy constraint, but also explicitly incorporates the geometric coherence constraint associating multiple frames in a video, thus can naturally maintain the temporal coherence of the recovered video depths without introducing over-smoothing artifact. To make the inference problem tractable, we introduce an iterative optimization scheme by first initializing disparity maps using segmentation prior and then refining the disparities by means of bundle optimization. Unlike previous work estimating complex visibility parameters, our approach implicitly models the probabilistic visibility in a statistical way. The effectiveness of our automatic method is demonstrated using challenging video examples.

Although camera self-calibration and metric reconstruction have been extensively studied during the past decades, automatic metric reconstruction from long video sequences with varying focal length is still very challenging. Several critical issues in practical implementations are not adequately addressed. For example, how to select the initial frames for initializing the projective reconstruction? What criteria should be used? How to handle the large zooming problem? How to choose an appropriate moment for upgrading the projective reconstruction to a metric one? This paper gives a careful investigation of all these issues. Practical and effective approaches are proposed. In particular, we show that existing image-based distance is not an adequate measurement for selecting the initial frames. We propose a novel measurement to take into account the zoom degree, the self-calibration quality, as well as image-based distance. We then introduce a new strategy to decide when to upgrade the projective reconstruction to a metric one. Finally, to alleviate the heavy computational cost in the bundle adjustment, a local on-demand approach is proposed. Our method is also extensively compared with the state-of-the-art commercial software to evidence its robustness and stability.

This paper presents a new method to detect and accurately extract the moving object from a video sequence taken by a hand-held camera. In order to extract the high quality moving foreground, previous approaches usually assume that the background is static or through only planar-perspective transformation. In our method, based on the robust motion estimation, we are capable of handling challenging videos where the background contains complex depth and the camera undergoes unknown motions. We propose the appearance and structure consistency constraint in 3D warping to robustly model the background, which greatly improves the foreground separation even on the object boundary. The estimated dense motion field and the bi- layer segmentation result are iteratively refined where continuous and discrete optimizations are alternatively used. Experimental results of high quality moving object extraction from challenging videos demonstrate the effectiveness of our method.

This paper presents an automatic and robust approach to synthesize stereoscopic videos from ordinary monocular videos acquired by commodity video cameras. Instead of recovering the depth map, the proposed method synthesizes the binocular parallax in stereoscopic video directly from the motion parallax in monocular video. The synthesis is formulated as an optimization problem via introducing a cost function of the stereoscopic effects, the similarity, and the smoothness constraints. The optimization selects the most suitable frames in the input video for generating the stereoscopic video frames. With the optimized selection, convincing and smooth stereoscopic video can be synthesized even by simple constant-depth warping. No user interaction is required. We demonstrate the visually plausible results obtained given the input clips acquired by ordinary handheld video camera.

Abstract We present an efficient approach that merges the virtual objects into video sequences taken by a freely moving camera in a realistic manner. The composition is visually and geometrically consistent through three main steps. First, a robust camera tracking algorithm based on key frames is proposed, which precisely recovers the focal length with a novel multi‐frame strategy. Next, the concerned 3D models of the real scenes are reconstructed by means of an extended multi‐baseline algorithm. Finally, the virtual objects in the form of 3D models are integrated into the real scenes, with special cares on the interaction consistency including shadow casting, occlusions, and object animation. A variety of experiments have been implemented, which demonstrate the robustness and efficiency of our approach. Copyright © 2006 John Wiley & Sons, Ltd.