Kaixuan Wang

Wei Yin, Chi Zhang, Hao Chen et al. · tencent-ai

Reconstructing accurate 3D scenes from images is a long-standing vision task. Due to the ill-posedness of the single-image reconstruction problem, most well-established methods are built upon multi-view geometry. State-of-the-art (SOTA) monocular metric depth estimation methods can only handle a single camera model and are unable to perform mixed-data training due to the metric ambiguity. Meanwhile, SOTA monocular methods trained on large mixed datasets achieve zero-shot generalization by learning affine-invariant depths, which cannot recover real-world metrics. In this work, we show that the key to a zero-shot single-view metric depth model lies in the combination of large-scale data training and resolving the metric ambiguity from various camera models. We propose a canonical camera space transformation module, which explicitly addresses the ambiguity problems and can be effortlessly plugged into existing monocular models. Equipped with our module, monocular models can be stably trained with over 8 million images with thousands of camera models, resulting in zero-shot generalization to in-the-wild images with unseen camera settings. Experiments demonstrate SOTA performance of our method on 7 zero-shot benchmarks. Notably, our method won the championship in the 2nd Monocular Depth Estimation Challenge. Our method enables the accurate recovery of metric 3D structures on randomly collected internet images, paving the way for plausible single-image metrology. The potential benefits extend to downstream tasks, which can be significantly improved by simply plugging in our model. For example, our model relieves the scale drift issues of monocular-SLAM (Fig. 1), leading to high-quality metric scale dense mapping. The code is available at https://github.com/YvanYin/Metric3D.

Rui Li, Dong Gong, Wei Yin et al.

Multi-frame depth estimation generally achieves high accuracy relying on the multi-view geometric consistency. When applied in dynamic scenes, e.g., autonomous driving, this consistency is usually violated in the dynamic areas, leading to corrupted estimations. Many multi-frame methods handle dynamic areas by identifying them with explicit masks and compensating the multi-view cues with monocular cues represented as local monocular depth or features. The improvements are limited due to the uncontrolled quality of the masks and the underutilized benefits of the fusion of the two types of cues. In this paper, we propose a novel method to learn to fuse the multi-view and monocular cues encoded as volumes without needing the heuristically crafted masks. As unveiled in our analyses, the multi-view cues capture more accurate geometric information in static areas, and the monocular cues capture more useful contexts in dynamic areas. To let the geometric perception learned from multi-view cues in static areas propagate to the monocular representation in dynamic areas and let monocular cues enhance the representation of multi-view cost volume, we propose a cross-cue fusion (CCF) module, which includes the cross-cue attention (CCA) to encode the spatially non-local relative intra-relations from each source to enhance the representation of the other. Experiments on real-world datasets prove the significant effectiveness and generalization ability of the proposed method.

Jaime Spencer, C. Stella Qian, Michaela Trescakova et al.

This paper discusses the results for the second edition of the Monocular Depth Estimation Challenge (MDEC). This edition was open to methods using any form of supervision, including fully-supervised, self-supervised, multi-task or proxy depth. The challenge was based around the SYNS-Patches dataset, which features a wide diversity of environments with high-quality dense ground-truth. This includes complex natural environments, e.g. forests or fields, which are greatly underrepresented in current benchmarks. The challenge received eight unique submissions that outperformed the provided SotA baseline on any of the pointcloud- or image-based metrics. The top supervised submission improved relative F-Score by 27.62%, while the top self-supervised improved it by 16.61%. Supervised submissions generally leveraged large collections of datasets to improve data diversity. Self-supervised submissions instead updated the network architecture and pretrained backbones. These results represent a significant progress in the field, while highlighting avenues for future research, such as reducing interpolation artifacts at depth boundaries, improving self-supervised indoor performance and overall natural image accuracy.

Kai Cheng, Xiaoxiao Long, Wei Yin et al.

Multi-camera setups find widespread use across various applications, such as autonomous driving, as they greatly expand sensing capabilities. Despite the fast development of Neural radiance field (NeRF) techniques and their wide applications in both indoor and outdoor scenes, applying NeRF to multi-camera systems remains very challenging. This is primarily due to the inherent under-calibration issues in multi-camera setup, including inconsistent imaging effects stemming from separately calibrated image signal processing units in diverse cameras, and system errors arising from mechanical vibrations during driving that affect relative camera poses. In this paper, we present UC-NeRF, a novel method tailored for novel view synthesis in under-calibrated multi-view camera systems. Firstly, we propose a layer-based color correction to rectify the color inconsistency in different image regions. Second, we propose virtual warping to generate more viewpoint-diverse but color-consistent virtual views for color correction and 3D recovery. Finally, a spatiotemporally constrained pose refinement is designed for more robust and accurate pose calibration in multi-camera systems. Our method not only achieves state-of-the-art performance of novel view synthesis in multi-camera setups, but also effectively facilitates depth estimation in large-scale outdoor scenes with the synthesized novel views.

JunDa Cheng, Wei Yin, Kaixuan Wang et al.

Multi-view depth estimation has achieved impressive performance over various benchmarks. However, almost all current multi-view systems rely on given ideal camera poses, which are unavailable in many real-world scenarios, such as autonomous driving. In this work, we propose a new robustness benchmark to evaluate the depth estimation system under various noisy pose settings. Surprisingly, we find current multi-view depth estimation methods or single-view and multi-view fusion methods will fail when given noisy pose settings. To address this challenge, we propose a single-view and multi-view fused depth estimation system, which adaptively integrates high-confident multi-view and single-view results for both robust and accurate depth estimations. The adaptive fusion module performs fusion by dynamically selecting high-confidence regions between two branches based on a wrapping confidence map. Thus, the system tends to choose the more reliable branch when facing textureless scenes, inaccurate calibration, dynamic objects, and other degradation or challenging conditions. Our method outperforms state-of-the-art multi-view and fusion methods under robustness testing. Furthermore, we achieve state-of-the-art performance on challenging benchmarks (KITTI and DDAD) when given accurate pose estimations. Project website: https://github.com/Junda24/AFNet/.

Mu Hu, Wei Yin, Chi Zhang et al.

We introduce Metric3D v2, a geometric foundation model for zero-shot metric depth and surface normal estimation from a single image, which is crucial for metric 3D recovery. While depth and normal are geometrically related and highly complimentary, they present distinct challenges. SoTA monocular depth methods achieve zero-shot generalization by learning affine-invariant depths, which cannot recover real-world metrics. Meanwhile, SoTA normal estimation methods have limited zero-shot performance due to the lack of large-scale labeled data. To tackle these issues, we propose solutions for both metric depth estimation and surface normal estimation. For metric depth estimation, we show that the key to a zero-shot single-view model lies in resolving the metric ambiguity from various camera models and large-scale data training. We propose a canonical camera space transformation module, which explicitly addresses the ambiguity problem and can be effortlessly plugged into existing monocular models. For surface normal estimation, we propose a joint depth-normal optimization module to distill diverse data knowledge from metric depth, enabling normal estimators to learn beyond normal labels. Equipped with these modules, our depth-normal models can be stably trained with over 16 million of images from thousands of camera models with different-type annotations, resulting in zero-shot generalization to in-the-wild images with unseen camera settings. Our method enables the accurate recovery of metric 3D structures on randomly collected internet images, paving the way for plausible single-image metrology. Our project page is at https://JUGGHM.github.io/Metric3Dv2.

Kaixuan Wang, Fei Gao, Shaojie Shen

In this paper, we propose a novel dense surfel mapping system that scales well in different environments with only CPU computation. Using a sparse SLAM system to estimate camera poses, the proposed mapping system can fuse intensity images and depth images into a globally consistent model. The system is carefully designed so that it can build from room-scale environments to urban-scale environments using depth images from RGB-D cameras, stereo cameras or even a monocular camera. First, superpixels extracted from both intensity and depth images are used to model surfels in the system. superpixel-based surfels make our method both run-time efficient and memory efficient. Second, surfels are further organized according to the pose graph of the SLAM system to achieve $O(1)$ fusion time regardless of the scale of reconstructed models. Third, a fast map deformation using the optimized pose graph enables the map to achieve global consistency in real-time. The proposed surfel mapping system is compared with other state-of-the-art methods on synthetic datasets. The performances of urban-scale and room-scale reconstruction are demonstrated using the KITTI dataset and autonomous aggressive flights, respectively. The code is available for the benefit of the community.

Xuelun Shen, Zhipeng Cai, Wei Yin et al.

Image matching is a fundamental computer vision problem. While learning-based methods achieve state-of-the-art performance on existing benchmarks, they generalize poorly to in-the-wild images. Such methods typically need to train separate models for different scene types and are impractical when the scene type is unknown in advance. One of the underlying problems is the limited scalability of existing data construction pipelines, which limits the diversity of standard image matching datasets. To address this problem, we propose GIM, a self-training framework for learning a single generalizable model based on any image matching architecture using internet videos, an abundant and diverse data source. Given an architecture, GIM first trains it on standard domain-specific datasets and then combines it with complementary matching methods to create dense labels on nearby frames of novel videos. These labels are filtered by robust fitting, and then enhanced by propagating them to distant frames. The final model is trained on propagated data with strong augmentations. We also propose ZEB, the first zero-shot evaluation benchmark for image matching. By mixing data from diverse domains, ZEB can thoroughly assess the cross-domain generalization performance of different methods. Applying GIM consistently improves the zero-shot performance of 3 state-of-the-art image matching architectures; with 50 hours of YouTube videos, the relative zero-shot performance improves by 8.4%-18.1%. GIM also enables generalization to extreme cross-domain data such as Bird Eye View (BEV) images of projected 3D point clouds (Fig. 1(c)). More importantly, our single zero-shot model consistently outperforms domain-specific baselines when evaluated on downstream tasks inherent to their respective domains. The video presentation is available at https://www.youtube.com/watch?v=FU_MJLD8LeY.

Kai Luo, Yakun Ju, Lin Qi et al.

Predicting accurate normal maps of objects from two-dimensional images in regions of complex structure and spatial material variations is challenging using photometric stereo methods due to the influence of surface reflection properties caused by variations in object geometry and surface materials. To address this issue, we propose a photometric stereo network called a RMAFF-PSN that uses residual multiscale attentional feature fusion to handle the ``difficult'' regions of the object. Unlike previous approaches that only use stacked convolutional layers to extract deep features from the input image, our method integrates feature information from different resolution stages and scales of the image. This approach preserves more physical information, such as texture and geometry of the object in complex regions, through shallow-deep stage feature extraction, double branching enhancement, and attention optimization. To test the network structure under real-world conditions, we propose a new real dataset called Simple PS data, which contains multiple objects with varying structures and materials. Experimental results on a publicly available benchmark dataset demonstrate that our method outperforms most existing calibrated photometric stereo methods for the same number of input images, especially in the case of highly non-convex object structures. Our method also obtains good results under sparse lighting conditions.

Kaixuan Wang, Jason T. Jacques, Chenxin Diao et al.

Access to accurate and actionable harm reduction information can directly impact the health outcomes of People Who Use Drugs (PWUD), yet existing online channels often fail to meet their diverse and dynamic needs due to limitations in adaptability, accessibility, and the pervasive impact of stigma. Large Language Models (LLMs) present a novel opportunity to enhance information provision, but their application in such a high-stakes domain is under-explored and presents socio-technical challenges. This paper investigates how LLMs can be responsibly designed to support the information needs of PWUD. Through a qualitative workshop involving diverse stakeholder groups (academics, harm reduction practitioners, and an online community moderator), we explored LLM capabilities, identified potential use cases, and delineated core design considerations. Our findings reveal that while LLMs can address some existing information barriers (e.g., by offering responsive, multilingual, and potentially less stigmatising interactions), their effectiveness is contingent upon overcoming challenges related to ethical alignment with harm reduction principles, nuanced contextual understanding, effective communication, and clearly defined operational boundaries. We articulate design pathways emphasising collaborative co-design with experts and PWUD to develop LLM systems that are helpful, safe, and responsibly governed. This work contributes empirically grounded insights and actionable design considerations for the responsible development of LLMs as supportive tools within the harm reduction ecosystem.

Kechun Xu, Xunlong Xia, Kaixuan Wang et al.

We study the task of language-conditioned pick and place in clutter, where a robot should grasp a target object in open clutter and move it to a specified place. Some approaches learn end-to-end policies with features from vision foundation models, requiring large datasets. Others combine foundation models in a zero-shot setting, suffering from cascading errors. In addition, they primarily leverage vision and language foundation models, focusing less on action priors. In this paper, we aim to develop an effective policy by integrating foundation priors from vision, language, and action. We propose A$^2$, an action prior alignment method that aligns unconditioned action priors with 3D vision-language priors by learning one attention layer. The alignment formulation enables our policy to train with less data and preserve zero-shot generalization capabilities. We show that a shared policy for both pick and place actions enhances the performance for each task, and introduce a policy adaptation scheme to accommodate the multi-modal nature of actions. Extensive experiments in simulation and the real-world show that our policy achieves higher task success rates with fewer steps for both pick and place tasks in clutter, effectively generalizing to unseen objects and language instructions. Videos and codes are available at https://xukechun.github.io/papers/A2.

Zhongliang Guo, Junhao Dong, Yifei Qian et al.

Neural style transfer (NST) generates new images by combining the style of one image with the content of another. However, unauthorized NST can exploit artwork, raising concerns about artists' rights and motivating the development of proactive protection methods. We propose Locally Adaptive Adversarial Color Attack (LAACA), empowering artists to protect their artwork from unauthorized style transfer by processing before public release. By delving into the intricacies of human visual perception and the role of different frequency components, our method strategically introduces frequency-adaptive perturbations in the image. These perturbations significantly degrade the generation quality of NST while maintaining an acceptable level of visual change in the original image, ensuring that potential infringers are discouraged from using the protected artworks, because of its bad NST generation quality. Additionally, existing metrics often overlook the importance of color fidelity in evaluating color-mattered tasks, such as the quality of NST-generated images, which is crucial in the context of artistic works. To comprehensively assess the color-mattered tasks, we propose the Adversarial Color Distance Metric (ACDM), designed to quantify the color difference of images pre- and post-manipulations. Experimental results confirm that attacking NST using LAACA results in visually inferior style transfer, and the ACDM can efficiently measure color-mattered tasks. By providing artists with a tool to safeguard their intellectual property, our work relieves the socio-technical challenges posed by the misuse of NST in the art community.

Ke Wang, Kaixuan Wang, Shaojie Shen

Many approaches have been proposed to estimate camera poses by directly minimizing photometric error. However, due to the non-convex property of direct alignment, proper initialization is still required for these methods. Many robust norms (e.g. Huber norm) have been proposed to deal with the outlier terms caused by incorrect initializations. These robust norms are solely defined on the magnitude of each error term. In this paper, we propose a novel robust norm, named FlowNorm, that exploits the information from both the local error term and the global image registration information. While the local information is defined on patch alignments, the global information is estimated using a learning-based network. Using both the local and global information, we achieve an unprecedented convergence range in which images can be aligned given large view angle changes or small overlaps. We further demonstrate the usability of the proposed robust norm by integrating it into the direct methods DSO and BA-Net, and generate more robust and accurate results in real-time.

Kaixuan Wang, Shaojie Shen

We propose a learning-based method that solves monocular stereo and can be extended to fuse depth information from multiple target frames. Given two unconstrained images from a monocular camera with known intrinsic calibration, our network estimates relative camera poses and the depth map of the source image. The core contribution of the proposed method is threefold. First, a network is tailored for static scenes that jointly estimates the optical flow and camera motion. By the joint estimation, the optical flow search space is gradually reduced resulting in an efficient and accurate flow estimation. Second, a novel triangulation layer is proposed to encode the estimated optical flow and camera motion while avoiding common numerical issues caused by epipolar. Third, beyond two-view depth estimation, we further extend the above networks to fuse depth information from multiple target images and estimate the depth map of the source image. To further benefit the research community, we introduce tools to generate photorealistic structure-from-motion datasets such that deep networks can be well trained and evaluated. The proposed method is compared with previous methods and achieves state-of-the-art results within less time. Images from real-world applications and Google Earth are used to demonstrate the generalization ability of the method.

Wenchao Ding, Wenliang Gao, Kaixuan Wang et al.

Trajectory replanning for quadrotors is essential to enable fully autonomous flight in unknown environments. Hierarchical motion planning frameworks, which combine path planning with path parameterization, are popular due to their time efficiency. However, the path planning cannot properly deal with non-static initial states of the quadrotor, which may result in non-smooth or even dynamically infeasible trajectories. In this paper, we present an efficient kinodynamic replanning framework by exploiting the advantageous properties of the B-spline, which facilitates dealing with the non-static state and guarantees safety and dynamical feasibility. Our framework starts with an efficient B-spline-based kinodynamic (EBK) search algorithm which finds a feasible trajectory with minimum control effort and time. To compensate for the discretization induced by the EBK search, an elastic optimization (EO) approach is proposed to refine the control point placement to the optimal location. Systematic comparisons against the state-of-the-art are conducted to validate the performance. Comprehensive onboard experiments using two different vision-based quadrotors are carried out showing the general applicability of the framework.

Yonggen Ling, Kaixuan Wang, Shaojie Shen

This paper presents a probabilistic approach for online dense reconstruction using a single monocular camera moving through the environment. Compared to spatial stereo, depth estimation from motion stereo is challenging due to insufficient parallaxes, visual scale changes, pose errors, etc. We utilize both the spatial and temporal correlations of consecutive depth estimates to increase the robustness and accuracy of monocular depth estimation. An online, recursive, probabilistic scheme to compute depth estimates, with corresponding covariances and inlier probability expectations, is proposed in this work. We integrate the obtained depth hypotheses into dense 3D models in an uncertainty-aware way. We show the effectiveness and efficiency of our proposed approach by comparing it with state-of-the-art methods in the TUM RGB-D SLAM and ICL-NUIM dataset. Online indoor and outdoor experiments are also presented for performance demonstration.

Wenchao Ding, Wenliang Gao, Kaixuan Wang et al.

We focus on a replanning scenario for quadrotors where considering time efficiency, non-static initial state and dynamical feasibility is of great significance. We propose a real-time B-spline based kinodynamic (RBK) search algorithm, which transforms a position-only shortest path search (such as A* and Dijkstra) into an efficient kinodynamic search, by exploring the properties of B-spline parameterization. The RBK search is greedy and produces a dynamically feasible time-parameterized trajectory efficiently, which facilitates non-static initial state of the quadrotor. To cope with the limitation of the greedy search and the discretization induced by a grid structure, we adopt an elastic optimization (EO) approach as a post-optimization process, to refine the control point placement provided by the RBK search. The EO approach finds the optimal control point placement inside an expanded elastic tube which represents the free space, by solving a Quadratically Constrained Quadratic Programming (QCQP) problem. We design a receding horizon replanner based on the local control property of B-spline. A systematic comparison of our method against two state-of-the-art methods is provided. We integrate our replanning system with a monocular vision-based quadrotor and validate our performance onboard.

Kaixuan Wang, Shaojie Shen

Although deep neural networks have been widely applied to computer vision problems, extending them into multiview depth estimation is non-trivial. In this paper, we present MVDepthNet, a convolutional network to solve the depth estimation problem given several image-pose pairs from a localized monocular camera in neighbor viewpoints. Multiview observations are encoded in a cost volume and then combined with the reference image to estimate the depth map using an encoder-decoder network. By encoding the information from multiview observations into the cost volume, our method achieves real-time performance and the flexibility of traditional methods that can be applied regardless of the camera intrinsic parameters and the number of images. Geometric data augmentation is used to train MVDepthNet. We further apply MVDepthNet in a monocular dense mapping system that continuously estimates depth maps using a single localized moving camera. Experiments show that our method can generate depth maps efficiently and precisely.