Weihao Yuan

Jiaxin Shi, Xidong Zhang, Fucai Zhu et al.

We propose a 3D-thinking-guided co-training framework that enables vision-language-action (VLA) models to perform 3D spatial reasoning implicitly during action prediction. Our core insight is that 3D geometry perception and 3D spatial reasoning are distinct capabilities that can be disentangled and injected at different feature hierarchies. During training, three tightly coupled components work in concert primarily within the latent space: (1) To gain geometric priors, a latent 3D geometry perception module aligns intermediate visual features with a 3D foundation model, acquiring low-level geometric cues without architectural modifications to the VLM backbone. (2) Complementing this, an online 3D reasoning distillation module mitigates the prompt-induced reasoning gap via a shared reasoning anchor token. During 3D VLM co-training, this anchor is emitted as the first output token to robustly encode spatial priors. During VLA training, it serves as an input token inserted between the task and action instructions, transferring high-level spatial thinking from explicit teacher reasoning prompts to student action prompts without chain-of-thought text generation. (3) These disentangled geometric and reasoning features are then united by a spatially augmented action integration, which jointly injects them into the action-query tokens as hierarchical spatial conditions to prevent action shortcuts. At deployment, our method retains only its lightweight adapters to perform implicit 3D reasoning, discarding the 3D foundation model and the teacher branch used for supervision. Consequently, it operates purely on 2D images without 3D sensors, external models, or explicit text generation while preventing catastrophic forgetting of the pretrained VLM, achieving state-of-the-art performance on LIBERO, LIBERO-PLUS, SimplerEnv, and real-world manipulation tasks.

Meng Li, Senbo Wang, Weihao Yuan et al.

Monocular depth estimation is an ambiguous problem, thus global structural cues play an important role in current data-driven single-view depth estimation methods. Panorama images capture the complete spatial information of their surroundings utilizing the equirectangular projection which introduces large distortion. This requires the depth estimation method to be able to handle the distortion and extract global context information from the image. In this paper, we propose an end-to-end deep network for monocular panorama depth estimation on a unit spherical surface. Specifically, we project the feature maps extracted from equirectangular images onto unit spherical surface sampled by uniformly distributed grids, where the decoder network can aggregate the information from the distortion-reduced feature maps. Meanwhile, we propose a global cross-attention-based fusion module to fuse the feature maps from skip connection and enhance the ability to obtain global context. Experiments are conducted on five panorama depth estimation datasets, and the results demonstrate that the proposed method substantially outperforms previous state-of-the-art methods. All related codes will be open-sourced in the upcoming days.

Weihao Yuan, Xiaodong Gu, Zuozhuo Dai et al.

Estimating the accurate depth from a single image is challenging since it is inherently ambiguous and ill-posed. While recent works design increasingly complicated and powerful networks to directly regress the depth map, we take the path of CRFs optimization. Due to the expensive computation, CRFs are usually performed between neighborhoods rather than the whole graph. To leverage the potential of fully-connected CRFs, we split the input into windows and perform the FC-CRFs optimization within each window, which reduces the computation complexity and makes FC-CRFs feasible. To better capture the relationships between nodes in the graph, we exploit the multi-head attention mechanism to compute a multi-head potential function, which is fed to the networks to output an optimized depth map. Then we build a bottom-up-top-down structure, where this neural window FC-CRFs module serves as the decoder, and a vision transformer serves as the encoder. The experiments demonstrate that our method significantly improves the performance across all metrics on both the KITTI and NYUv2 datasets, compared to previous methods. Furthermore, the proposed method can be directly applied to panorama images and outperforms all previous panorama methods on the MatterPort3D dataset. Project page: https://weihaosky.github.io/newcrfs.

Weihao Yuan, Weichao Shen, Yisheng He et al.

Motion generation from discrete quantization offers many advantages over continuous regression, but at the cost of inevitable approximation errors. Previous methods usually quantize the entire body pose into one code, which not only faces the difficulty in encoding all joints within one vector but also loses the spatial relationship between different joints. Differently, in this work we quantize each individual joint into one vector, which i) simplifies the quantization process as the complexity associated with a single joint is markedly lower than that of the entire pose; ii) maintains a spatial-temporal structure that preserves both the spatial relationships among joints and the temporal movement patterns; iii) yields a 2D token map, which enables the application of various 2D operations widely used in 2D images. Grounded in the 2D motion quantization, we build a spatial-temporal modeling framework, where 2D joint VQVAE, temporal-spatial 2D masking technique, and spatial-temporal 2D attention are proposed to take advantage of spatial-temporal signals among the 2D tokens. Extensive experiments demonstrate that our method significantly outperforms previous methods across different datasets, with a 26.6% decrease of FID on HumanML3D and a 29.9% decrease on KIT-ML. Project page: https://aigc3d.github.io/mogents.

Lingteng Qiu, Guanying Chen, Xiaodong Gu et al.

Lifting 2D diffusion for 3D generation is a challenging problem due to the lack of geometric prior and the complex entanglement of materials and lighting in natural images. Existing methods have shown promise by first creating the geometry through score-distillation sampling (SDS) applied to rendered surface normals, followed by appearance modeling. However, relying on a 2D RGB diffusion model to optimize surface normals is suboptimal due to the distribution discrepancy between natural images and normals maps, leading to instability in optimization. In this paper, recognizing that the normal and depth information effectively describe scene geometry and be automatically estimated from images, we propose to learn a generalizable Normal-Depth diffusion model for 3D generation. We achieve this by training on the large-scale LAION dataset together with the generalizable image-to-depth and normal prior models. In an attempt to alleviate the mixed illumination effects in the generated materials, we introduce an albedo diffusion model to impose data-driven constraints on the albedo component. Our experiments show that when integrated into existing text-to-3D pipelines, our models significantly enhance the detail richness, achieving state-of-the-art results. Our project page is https://aigc3d.github.io/richdreamer/.

Heng Li, Xiaodong Gu, Weihao Yuan et al.

There is an emerging trend of using neural implicit functions for map representation in Simultaneous Localization and Mapping (SLAM). Some pioneer works have achieved encouraging results on RGB-D SLAM. In this paper, we present a dense RGB SLAM method with neural implicit map representation. To reach this challenging goal without depth input, we introduce a hierarchical feature volume to facilitate the implicit map decoder. This design effectively fuses shape cues across different scales to facilitate map reconstruction. Our method simultaneously solves the camera motion and the neural implicit map by matching the rendered and input video frames. To facilitate optimization, we further propose a photometric warping loss in the spirit of multi-view stereo to better constrain the camera pose and scene geometry. We evaluate our method on commonly used benchmarks and compare it with modern RGB and RGB-D SLAM systems. Our method achieves favorable results than previous methods and even surpasses some recent RGB-D SLAM methods.The code is at poptree.github.io/DIM-SLAM/.

Weihao Yuan, Xiaodong Gu, Heng Li et al.

Monocular scene reconstruction from posed images is challenging due to the complexity of a large environment. Recent volumetric methods learn to directly predict the TSDF volume and have demonstrated promising results in this task. However, most methods focus on how to extract and fuse the 2D features to a 3D feature volume, but none of them improve the way how the 3D volume is aggregated. In this work, we propose an SDF transformer network, which replaces the role of 3D CNN for better 3D feature aggregation. To reduce the explosive computation complexity of the 3D multi-head attention, we propose a sparse window attention module, where the attention is only calculated between the non-empty voxels within a local window. Then a top-down-bottom-up 3D attention network is built for 3D feature aggregation, where a dilate-attention structure is proposed to prevent geometry degeneration, and two global modules are employed to equip with global receptive fields. The experiments on multiple datasets show that this 3D transformer network generates a more accurate and complete reconstruction, which outperforms previous methods by a large margin. Remarkably, the mesh accuracy is improved by 41.8%, and the mesh completeness is improved by 25.3% on the ScanNet dataset. Project page: https://weihaosky.github.io/sdfformer.

Xiaodong Gu, Chengzhou Tang, Weihao Yuan et al.

3D motion estimation including scene flow and point cloud registration has drawn increasing interest. Inspired by 2D flow estimation, recent methods employ deep neural networks to construct the cost volume for estimating accurate 3D flow. However, these methods are limited by the fact that it is difficult to define a search window on point clouds because of the irregular data structure. In this paper, we avoid this irregularity by a simple yet effective method.We decompose the problem into two interlaced stages, where the 3D flows are optimized point-wisely at the first stage and then globally regularized in a recurrent network at the second stage. Therefore, the recurrent network only receives the regular point-wise information as the input. In the experiments, we evaluate the proposed method on both the 3D scene flow estimation and the point cloud registration task. For 3D scene flow estimation, we make comparisons on the widely used FlyingThings3D and KITTIdatasets. For point cloud registration, we follow previous works and evaluate the data pairs with large pose and partially overlapping from ModelNet40. The results show that our method outperforms the previous method and achieves a new state-of-the-art performance on both 3D scene flow estimation and point cloud registration, which demonstrates the superiority of the proposed zero-order method on irregular point cloud data.

Yingxin Lai, Yafei Zhou, Fucai Zhu et al.

While rule-based reinforcement learning has recently catalyzed explicit reasoning in multimodal models, tactile reasoning remains largely underexplored. Existing tactile-language models primarily rely on supervised or contrastive objectives, which limits their capacity to ground predictions in physical evidence or rectify misleading visual priors. Tactile reasoning introduces two modality-specific challenges: the ordinal nature of physical attributes (e.g., hardness, roughness) and the cross-sensor distribution shifts inherent in optical tactile hardware. In this work, we introduce TouchReason-1M, a large-scale multimodal dataset comprising over 1M synchronized tactile pairs across four distinct sensors, and TouchReason-Bench, a rigorous framework for evaluating tactile perception and visual-tactile conflict resolution. Building upon these, we propose Touch-R1, a tactile reasoning MLLM based on Qwen2.5-VL-7B. Touch-R1 is trained via a tactile-grounded GRPO objective that combines ordinal-aware accuracy, cross-sensor physical consistency, structured-format control, and an input-side tactile grounding objective. Specifically, the tactile-use reward assigns credit only when authentic tactile inputs yield superior correctness relative to counterfactual controls where the tactile stream is removed, shuffled, or noise-masked. On TouchReason-Bench, Touch-R1-7B outperforms Octopi-13B by 18.4\% and GPT-4o by 24.7\% on average. Its structured reasoning traces reveal emergent behaviors of probing, comparison, and revision, demonstrating that R1-style reasoning can be effectively grounded in physical contact.

Xiaodong Gu, Weihao Yuan, Heng Li et al.

Neural implicit surface reconstruction has become a new trend in reconstructing a detailed 3D shape from images. In previous methods, however, the 3D scene is only encoded by the MLPs which do not have an explicit 3D structure. To better represent 3D shapes, we introduce a volume encoding to explicitly encode the spatial information. We further design hierarchical volumes to encode the scene structures in multiple scales. The high-resolution volumes capture the high-frequency geometry details since spatially varying features could be learned from different 3D points, while the low-resolution volumes enforce the spatial consistency to keep the shape smooth since adjacent locations possess the same low-resolution feature. In addition, we adopt a sparse structure to reduce the memory consumption at high-resolution volumes, and two regularization terms to enhance results smoothness. This hierarchical volume encoding could be appended to any implicit surface reconstruction method as a plug-and-play module, and can generate a smooth and clean reconstruction with more details. Superior performance is demonstrated in DTU, EPFL, and BlendedMVS datasets with significant improvement on the standard metrics.

Fan Yang, Heyuan Li, Peihao Li et al.

Generating high-fidelity upper-body 3D avatars from one-shot input image remains a significant challenge. Current 3D avatar generation methods, which rely on large reconstruction models, are fast and capable of producing stable body structures, but they often suffer from artifacts such as blurry textures and stiff, unnatural motion. In contrast, generative video models show promising performance by synthesizing photorealistic and dynamic results, but they frequently struggle with unstable behavior, including body structural errors and identity drift. To address these limitations, we propose a novel approach that combines the strengths of both paradigms. Our framework employs a 3D reconstruction model to provide robust structural and appearance priors, which in turn guides a real-time autoregressive video diffusion model for rendering. This process enables the model to synthesize high-frequency, photorealistic details and fluid dynamics in real time, effectively reducing texture blur and motion stiffness while preventing the structural inconsistencies common in video generation methods. By uniting the geometric stability of 3D reconstruction with the generative capabilities of video models, our method produces high-fidelity digital avatars with realistic appearance and dynamic, temporally coherent motion. Experiments demonstrate that our approach significantly reduces artifacts and achieves substantial improvements in visual quality over leading methods, providing a robust and efficient solution for real-time applications such as gaming and virtual reality. Project page: https://lhyfst.github.io/visa

Yichi Zhang, Weihao Yuan, Yizhuo Zhang et al.

Vision-Language-Action (VLA) models improve action generation by conditioning policies on rich vision-language information. However, current auto-regressive policies are constrained by three bottlenecks: (1) architectural bias drives models to overlook visual details, (2) an excessive number of visual tokens makes attention difficult to focus on the correct regions, and (3) task-irrelevant visual information introduces substantial noise - together severely impairing the quality of action. In this paper, we investigate how to effectively utilize different visual representations for action generation. To this end, we first empirically validate the above issues and show that VLA performance is primarily limited by how visual information is utilized, rather than by the quality of visual representations. Based on these insights, we introduce FocusVLA, a novel paradigm that directs the model's attention to task-relevant visual regions to effectively bridge vision to action. Specifically, we first propose Modality Cascaded Attention to eliminate shortcut pathways, thereby compelling VLA models to rely on task-relevant visual details for action generation. Furthermore, we propose Focus Attention, which dynamically selects task-relevant visual patches to control information quantity while explicitly modulating their influence to suppress task-irrelevant noise. Extensive experiments on both simulated and real-world robotic benchmarks demonstrate that FocusVLA not only effectively leverages visual details to perform dexterous manipulations, but also substantially improves performance and accelerates convergence across a variety of tasks.

Yisheng He, Xiaodong Gu, Xiaodan Ye et al.

We present LAM, an innovative Large Avatar Model for animatable Gaussian head reconstruction from a single image. Unlike previous methods that require extensive training on captured video sequences or rely on auxiliary neural networks for animation and rendering during inference, our approach generates Gaussian heads that are immediately animatable and renderable. Specifically, LAM creates an animatable Gaussian head in a single forward pass, enabling reenactment and rendering without additional networks or post-processing steps. This capability allows for seamless integration into existing rendering pipelines, ensuring real-time animation and rendering across a wide range of platforms, including mobile phones. The centerpiece of our framework is the canonical Gaussian attributes generator, which utilizes FLAME canonical points as queries. These points interact with multi-scale image features through a Transformer to accurately predict Gaussian attributes in the canonical space. The reconstructed canonical Gaussian avatar can then be animated utilizing standard linear blend skinning (LBS) with corrective blendshapes as the FLAME model did and rendered in real-time on various platforms. Our experimental results demonstrate that LAM outperforms state-of-the-art methods on existing benchmarks. Our code and video are available at https://aigc3d.github.io/projects/LAM/

Baoyou Chen, Hanchen Xia, Peng Tu et al.

Autoregressive vision-language models (VLMs) deliver strong multimodal capability, but their token-by-token decoding imposes a fundamental inference bottleneck. Diffusion VLMs offer a more parallel decoding paradigm, yet directly converting a pretrained autoregressive VLM into a large-block diffusion VLM (dVLM) often leads to substantial quality degradation. In this work, we present BARD, a simple and effective bridging framework that converts a pretrained autoregressive VLM into a same-architecture, decoding-efficient dVLM. Our approach combines progressive supervised block merging, which gradually enlarges the decoding block size, with stage-wise intra-dVLM distillation from a fixed small-block diffusion anchor to recover performance lost at larger blocks. We further incorporate a mixed noise scheduler to improve robustness and token revision during denoising, and memory-friendly training to enable efficient training on long multimodal sequences. A key empirical finding is that direct autoregressive-to-diffusion distillation is poorly aligned and can even hurt performance, whereas distillation within the diffusion regime is consistently effective. Experimental results show that, with $\leq 4.4M$ data, BARD-VL transfers strong multimodal capability from Qwen3-VL to a large-block dVLM. Remarkably, BARD-VL establishes a new SOTA among comparable-scale open dVLMs on our evaluation suite at both 4B and 8B scales. At the same time, BARD-VL achieves up to \textbf{3$\times$} decoding throughput speedup compared to the source model.

Yan Chen, Hanlin Shang, Ce Liu et al.

Video face restoration faces a critical challenge in maintaining temporal consistency while recovering fine facial details from degraded inputs. This paper presents a novel approach that extends Vector-Quantized Variational Autoencoders (VQ-VAEs), pretrained on static high-quality portraits, into a video restoration framework through variational latent space modeling. Our key innovation lies in reformulating discrete codebook representations as Dirichlet-distributed continuous variables, enabling probabilistic transitions between facial features across frames. A spatio-temporal Transformer architecture jointly models inter-frame dependencies and predicts latent distributions, while a Laplacian-constrained reconstruction loss combined with perceptual (LPIPS) regularization enhances both pixel accuracy and visual quality. Comprehensive evaluations on blind face restoration, video inpainting, and facial colorization tasks demonstrate state-of-the-art performance. This work establishes an effective paradigm for adapting intensive image priors, pretrained on high-quality images, to video restoration while addressing the critical challenge of flicker artifacts. The source code has been open-sourced and is available at https://github.com/fudan-generative-vision/DicFace.

Xiaodong Gu, Weihao Yuan, Zuozhuo Dai et al.

There are increasing interests of studying the video-to-depth (V2D) problem with machine learning techniques. While earlier methods directly learn a mapping from images to depth maps and camera poses, more recent works enforce multi-view geometry constraints through optimization embedded in the learning framework. This paper presents a novel optimization method based on recurrent neural networks to further exploit the potential of neural networks in V2D. Specifically, our neural optimizer alternately updates the depth and camera poses through iterations to minimize a feature-metric cost, and two gated recurrent units iteratively improve the results by tracing historical information. Extensive experimental results demonstrate that our method outperforms previous methods and is more efficient in computation and memory consumption than cost-volume-based methods. In particular, our self-supervised method outperforms previous supervised methods on the KITTI and ScanNet datasets. Our source code is available at https://github.com/aliyun/dro-sfm.

Zuozhuo Dai, Guangyuan Wang, Weihao Yuan et al.

State-of-the-art unsupervised re-ID methods train the neural networks using a memory-based non-parametric softmax loss. Instance feature vectors stored in memory are assigned pseudo-labels by clustering and updated at instance level. However, the varying cluster sizes leads to inconsistency in the updating progress of each cluster. To solve this problem, we present Cluster Contrast which stores feature vectors and computes contrast loss at the cluster level. Our approach employs a unique cluster representation to describe each cluster, resulting in a cluster-level memory dictionary. In this way, the consistency of clustering can be effectively maintained throughout the pipline and the GPU memory consumption can be significantly reduced. Thus, our method can solve the problem of cluster inconsistency and be applicable to larger data sets. In addition, we adopt different clustering algorithms to demonstrate the robustness and generalization of our framework. The application of Cluster Contrast to a standard unsupervised re-ID pipeline achieves considerable improvements of 9.9%, 8.3%, 12.1% compared to state-of-the-art purely unsupervised re-ID methods and 5.5%, 4.8%, 4.4% mAP compared to the state-of-the-art unsupervised domain adaptation re-ID methods on the Market, Duke, and MSMT17 datasets. Code is available at https://github.com/alibaba/cluster-contrast.

Qi Zuo, Xiaodong Gu, Lingteng Qiu et al.

Generating multi-view images based on text or single-image prompts is a critical capability for the creation of 3D content. Two fundamental questions on this topic are what data we use for training and how to ensure multi-view consistency. This paper introduces a novel framework that makes fundamental contributions to both questions. Unlike leveraging images from 2D diffusion models for training, we propose a dense consistent multi-view generation model that is fine-tuned from off-the-shelf video generative models. Images from video generative models are more suitable for multi-view generation because the underlying network architecture that generates them employs a temporal module to enforce frame consistency. Moreover, the video data sets used to train these models are abundant and diverse, leading to a reduced train-finetuning domain gap. To enhance multi-view consistency, we introduce a 3D-Aware Denoising Sampling, which first employs a feed-forward reconstruction module to get an explicit global 3D model, and then adopts a sampling strategy that effectively involves images rendered from the global 3D model into the denoising sampling loop to improve the multi-view consistency of the final images. As a by-product, this module also provides a fast way to create 3D assets represented by 3D Gaussians within a few seconds. Our approach can generate 24 dense views and converges much faster in training than state-of-the-art approaches (4 GPU hours versus many thousand GPU hours) with comparable visual quality and consistency. By further fine-tuning, our approach outperforms existing state-of-the-art methods in both quantitative metrics and visual effects. Our project page is aigc3d.github.io/VideoMV.

Lingteng Qiu, Xiaodong Gu, Peihao Li et al.

Animatable 3D human reconstruction from a single image is a challenging problem due to the ambiguity in decoupling geometry, appearance, and deformation. Recent advances in 3D human reconstruction mainly focus on static human modeling, and the reliance of using synthetic 3D scans for training limits their generalization ability. Conversely, optimization-based video methods achieve higher fidelity but demand controlled capture conditions and computationally intensive refinement processes. Motivated by the emergence of large reconstruction models for efficient static reconstruction, we propose LHM (Large Animatable Human Reconstruction Model) to infer high-fidelity avatars represented as 3D Gaussian splatting in a feed-forward pass. Our model leverages a multimodal transformer architecture to effectively encode the human body positional features and image features with attention mechanism, enabling detailed preservation of clothing geometry and texture. To further boost the face identity preservation and fine detail recovery, we propose a head feature pyramid encoding scheme to aggregate multi-scale features of the head regions. Extensive experiments demonstrate that our LHM generates plausible animatable human in seconds without post-processing for face and hands, outperforming existing methods in both reconstruction accuracy and generalization ability.

Lingteng Qiu, Shenhao Zhu, Qi Zuo et al.

Generating animatable human avatars from a single image is essential for various digital human modeling applications. Existing 3D reconstruction methods often struggle to capture fine details in animatable models, while generative approaches for controllable animation, though avoiding explicit 3D modeling, suffer from viewpoint inconsistencies in extreme poses and computational inefficiencies. In this paper, we address these challenges by leveraging the power of generative models to produce detailed multi-view canonical pose images, which help resolve ambiguities in animatable human reconstruction. We then propose a robust method for 3D reconstruction of inconsistent images, enabling real-time rendering during inference. Specifically, we adapt a transformer-based video generation model to generate multi-view canonical pose images and normal maps, pretraining on a large-scale video dataset to improve generalization. To handle view inconsistencies, we recast the reconstruction problem as a 4D task and introduce an efficient 3D modeling approach using 4D Gaussian Splatting. Experiments demonstrate that our method achieves photorealistic, real-time animation of 3D human avatars from in-the-wild images, showcasing its effectiveness and generalization capability.

Zhe Li, Yisheng He, Lei Zhong et al.

Generating motion sequences conforming to a target style while adhering to the given content prompts requires accommodating both the content and style. In existing methods, the information usually only flows from style to content, which may cause conflict between the style and content, harming the integration. Differently, in this work we build a bidirectional control flow between the style and the content, also adjusting the style towards the content, in which case the style-content collision is alleviated and the dynamics of the style is better preserved in the integration. Moreover, we extend the stylized motion generation from one modality, i.e. the style motion, to multiple modalities including texts and images through contrastive learning, leading to flexible style control on the motion generation. Extensive experiments demonstrate that our method significantly outperforms previous methods across different datasets, while also enabling multimodal signals control. The code of our method will be made publicly available.

Shenhao Zhu, Lingteng Qiu, Xiaodong Gu et al.

Existing 2D methods utilize UNet-based diffusion models to generate multi-view physically-based rendering (PBR) maps but struggle with multi-view inconsistency, while some 3D methods directly generate UV maps, encountering generalization issues due to the limited 3D data. To address these problems, we propose a two-stage approach, including multi-view generation and UV materials refinement. In the generation stage, we adopt a Diffusion Transformer (DiT) model to generate PBR materials, where both the specially designed multi-branch DiT and reference-based DiT blocks adopt a global attention mechanism to promote feature interaction and fusion between different views, thereby improving multi-view consistency. In addition, we adopt a PBR-based diffusion loss to ensure that the generated materials align with realistic physical principles. In the refinement stage, we propose a material-refined DiT that performs inpainting in empty areas and enhances details in UV space. Except for the normal condition, this refinement also takes the material map from the generation stage as an additional condition to reduce the learning difficulty and improve generalization. Extensive experiments show that our method achieves state-of-the-art performance in texturing 3D objects with PBR materials and provides significant advantages for graphics relighting applications. Project Page: https://lingtengqiu.github.io/2024/MCMat/

Yisheng He, Weihao Yuan, Siyu Zhu et al.

This paper enables high-fidelity, transferable NeRF editing by frequency decomposition. Recent NeRF editing pipelines lift 2D stylization results to 3D scenes while suffering from blurry results, and fail to capture detailed structures caused by the inconsistency between 2D editings. Our critical insight is that low-frequency components of images are more multiview-consistent after editing compared with their high-frequency parts. Moreover, the appearance style is mainly exhibited on the low-frequency components, and the content details especially reside in high-frequency parts. This motivates us to perform editing on low-frequency components, which results in high-fidelity edited scenes. In addition, the editing is performed in the low-frequency feature space, enabling stable intensity control and novel scene transfer. Comprehensive experiments conducted on photorealistic datasets demonstrate the superior performance of high-fidelity and transferable NeRF editing. The project page is at \url{https://aigc3d.github.io/freditor}.

Xiaoguang Han, Yushuang Wu, Luyue Shi et al.

MVImgNet is a large-scale dataset that contains multi-view images of ~220k real-world objects in 238 classes. As a counterpart of ImageNet, it introduces 3D visual signals via multi-view shooting, making a soft bridge between 2D and 3D vision. This paper constructs the MVImgNet2.0 dataset that expands MVImgNet into a total of ~520k objects and 515 categories, which derives a 3D dataset with a larger scale that is more comparable to ones in the 2D domain. In addition to the expanded dataset scale and category range, MVImgNet2.0 is of a higher quality than MVImgNet owing to four new features: (i) most shoots capture 360-degree views of the objects, which can support the learning of object reconstruction with completeness; (ii) the segmentation manner is advanced to produce foreground object masks of higher accuracy; (iii) a more powerful structure-from-motion method is adopted to derive the camera pose for each frame of a lower estimation error; (iv) higher-quality dense point clouds are reconstructed via advanced methods for objects captured in 360-degree views, which can serve for downstream applications. Extensive experiments confirm the value of the proposed MVImgNet2.0 in boosting the performance of large 3D reconstruction models. MVImgNet2.0 will be public at luyues.github.io/mvimgnet2, including multi-view images of all 520k objects, the reconstructed high-quality point clouds, and data annotation codes, hoping to inspire the broader vision community.

Chao He, Jianqiang Ren, Yuan Dong et al.

The 2D cartoon style is a prominent art form in digital character creation, particularly popular among younger audiences. While advancements in digital human technology have spurred extensive research into photorealistic digital humans and 3D characters, interactive 2D cartoon characters have received comparatively less attention. Unlike 3D counterparts, which require sophisticated construction and resource-intensive rendering, Live2D, a widely-used format for 2D cartoon characters, offers a more efficient alternative, which allows to animate 2D characters in a manner that simulates 3D movement without the necessity of building a complete 3D model. Furthermore, Live2D employs lightweight HTML5 (H5) rendering, improving both accessibility and efficiency. In this technical report, we introduce Textoon, an innovative method for generating diverse 2D cartoon characters in the Live2D format based on text descriptions. The Textoon leverages cutting-edge language and vision models to comprehend textual intentions and generate 2D appearance, capable of creating a wide variety of stunning and interactive 2D characters within one minute. The project homepage is https://human3daigc.github.io/Textoon_webpage/.

Zhe Li, Weihao Yuan, Weichao Shen et al.

Whole-body multi-modal human motion generation poses two primary challenges: creating an effective motion generation mechanism and integrating various modalities, such as text, speech, and music, into a cohesive framework. Unlike previous methods that usually employ discrete masked modeling or autoregressive modeling, we develop a continuous masked autoregressive motion transformer, where a causal attention is performed considering the sequential nature within the human motion. Within this transformer, we introduce a gated linear attention and an RMSNorm module, which drive the transformer to pay attention to the key actions and suppress the instability caused by either the abnormal movements or the heterogeneous distributions within multi-modalities. To further enhance both the motion generation and the multimodal generalization, we employ the DiT structure to diffuse the conditions from the transformer towards the targets. To fuse different modalities, AdaLN and cross-attention are leveraged to inject the text, speech, and music signals. Experimental results demonstrate that our framework outperforms previous methods across all modalities, including text-to-motion, speech-to-gesture, and music-to-dance. The code of our method will be made public.

Yingdong Hu, Yisheng He, Jinnan Chen et al.

Instant reconstruction of dynamic 3D humans from uncalibrated sparse-view videos is critical for numerous downstream applications. Existing methods, however, are either limited by the slow reconstruction speeds or incapable of generating novel-time representations. To address these challenges, we propose Forge4D, a feed-forward 4D human reconstruction and interpolation model that efficiently reconstructs temporally aligned representations from uncalibrated sparse-view videos, enabling both novel view and novel time synthesis. Our model simplifies the 4D reconstruction and interpolation problem as a joint task of streaming 3D Gaussian reconstruction and dense motion prediction. For the task of streaming 3D Gaussian reconstruction, we first reconstruct static 3D Gaussians from uncalibrated sparse-view images and then introduce learnable state tokens to enforce temporal consistency in a memory-friendly manner by interactively updating shared information across different timestamps. For novel time synthesis, we design a novel motion prediction module to predict dense motions for each 3D Gaussian between two adjacent frames, coupled with an occlusion-aware Gaussian fusion process to interpolate 3D Gaussians at arbitrary timestamps. To overcome the lack of the ground truth for dense motion supervision, we formulate dense motion prediction as a dense point matching task and introduce a self-supervised retargeting loss to optimize this module. An additional occlusion-aware optical flow loss is introduced to ensure motion consistency with plausible human movement, providing stronger regularization. Extensive experiments demonstrate the effectiveness of our model on both in-domain and out-of-domain datasets. Project page and code at: https://zhenliuzju.github.io/huyingdong/Forge4D.

Lingteng Qiu, Peihao Li, Qi Zuo et al.

Reconstructing an animatable 3D human from casually captured images of an articulated subject without camera or human pose information is a practical yet challenging task due to view misalignment, occlusions, and the absence of structural priors. While optimization-based methods can produce high-fidelity results from monocular or multi-view videos, they require accurate pose estimation and slow iterative optimization, limiting scalability in unconstrained scenarios. Recent feed-forward approaches enable efficient single-image reconstruction but struggle to effectively leverage multiple input images to reduce ambiguity and improve reconstruction accuracy. To address these challenges, we propose PF-LHM, a large human reconstruction model that generates high-quality 3D avatars in seconds from one or multiple casually captured pose-free images. Our approach introduces an efficient Encoder-Decoder Point-Image Transformer architecture, which fuses hierarchical geometric point features and multi-view image features through multimodal attention. The fused features are decoded to recover detailed geometry and appearance, represented using 3D Gaussian splats. Extensive experiments on both real and synthetic datasets demonstrate that our method unifies single- and multi-image 3D human reconstruction, achieving high-fidelity and animatable 3D human avatars without requiring camera and human pose annotations. Code and models will be released to the public.

Yushuang Wu, Luyue Shi, Junhao Cai et al.

Generalizable 3D object reconstruction from single-view RGB-D images remains a challenging task, particularly with real-world data. Current state-of-the-art methods develop Transformer-based implicit field learning, necessitating an intensive learning paradigm that requires dense query-supervision uniformly sampled throughout the entire space. We propose a novel approach, IPoD, which harmonizes implicit field learning with point diffusion. This approach treats the query points for implicit field learning as a noisy point cloud for iterative denoising, allowing for their dynamic adaptation to the target object shape. Such adaptive query points harness diffusion learning's capability for coarse shape recovery and also enhances the implicit representation's ability to delineate finer details. Besides, an additional self-conditioning mechanism is designed to use implicit predictions as the guidance of diffusion learning, leading to a cooperative system. Experiments conducted on the CO3D-v2 dataset affirm the superiority of IPoD, achieving 7.8% improvement in F-score and 28.6% in Chamfer distance over existing methods. The generalizability of IPoD is also demonstrated on the MVImgNet dataset. Our project page is at https://yushuang-wu.github.io/IPoD.

Zhengyi Zhao, Chen Song, Xiaodong Gu et al.

A fundamental problem in the texturing of 3D meshes using pre-trained text-to-image models is to ensure multi-view consistency. State-of-the-art approaches typically use diffusion models to aggregate multi-view inputs, where common issues are the blurriness caused by the averaging operation in the aggregation step or inconsistencies in local features. This paper introduces an optimization framework that proceeds in four stages to achieve multi-view consistency. Specifically, the first stage generates an over-complete set of 2D textures from a predefined set of viewpoints using an MV-consistent diffusion process. The second stage selects a subset of views that are mutually consistent while covering the underlying 3D model. We show how to achieve this goal by solving semi-definite programs. The third stage performs non-rigid alignment to align the selected views across overlapping regions. The fourth stage solves an MRF problem to associate each mesh face with a selected view. In particular, the third and fourth stages are iterated, with the cuts obtained in the fourth stage encouraging non-rigid alignment in the third stage to focus on regions close to the cuts. Experimental results show that our approach significantly outperforms baseline approaches both qualitatively and quantitatively. Project page: https://aigc3d.github.io/ConsistenTex.

Peng Li, Yisheng He, Yingdong Hu et al.

We present a feed-forward framework for Gaussian full-head synthesis from a single unposed image. Unlike previous work that relies on time-consuming GAN inversion and test-time optimization, our framework can reconstruct the Gaussian full-head model given a single unposed image in a single forward pass. This enables fast reconstruction and rendering during inference. To mitigate the lack of large-scale 3D head assets, we propose a large-scale synthetic dataset from trained 3D GANs and train our framework using only synthetic data. For efficient high-fidelity generation, we introduce a coarse-to-fine Gaussian head generation pipeline, where sparse points from the FLAME model interact with the image features by transformer blocks for feature extraction and coarse shape reconstruction, which are then densified for high-fidelity reconstruction. To fully leverage the prior knowledge residing in pretrained 3D GANs for effective reconstruction, we propose a dual-branch framework that effectively aggregates the structured spherical triplane feature and unstructured point-based features for more effective Gaussian head reconstruction. Experimental results show the effectiveness of our framework towards existing work. Project page at: https://panolam.github.io/.

Junhao Cai, Yuji Yang, Weihao Yuan et al.

This paper studies the problem of estimating physical properties (system identification) through visual observations. To facilitate geometry-aware guidance in physical property estimation, we introduce a novel hybrid framework that leverages 3D Gaussian representation to not only capture explicit shapes but also enable the simulated continuum to render object masks as 2D shape surrogates during training. We propose a new dynamic 3D Gaussian framework based on motion factorization to recover the object as 3D Gaussian point sets across different time states. Furthermore, we develop a coarse-to-fine filling strategy to generate the density fields of the object from the Gaussian reconstruction, allowing for the extraction of object continuums along with their surfaces and the integration of Gaussian attributes into these continuum. In addition to the extracted object surfaces, the Gaussian-informed continuum also enables the rendering of object masks during simulations, serving as 2D-shape guidance for physical property estimation. Extensive experimental evaluations demonstrate that our pipeline achieves state-of-the-art performance across multiple benchmarks and metrics. Additionally, we illustrate the effectiveness of the proposed method through real-world demonstrations, showcasing its practical utility. Our project page is at https://jukgei.github.io/project/gic.

Junhao Cai, Yisheng He, Weihao Yuan et al.

This paper studies a new open-set problem, the open-vocabulary category-level object pose and size estimation. Given human text descriptions of arbitrary novel object categories, the robot agent seeks to predict the position, orientation, and size of the target object in the observed scene image. To enable such generalizability, we first introduce OO3D-9D, a large-scale photorealistic dataset for this task. Derived from OmniObject3D, OO3D-9D is the largest and most diverse dataset in the field of category-level object pose and size estimation. It includes additional annotations for the symmetry axis of each category, which help resolve symmetric ambiguity. Apart from the large-scale dataset, we find another key to enabling such generalizability is leveraging the strong prior knowledge in pre-trained visual-language foundation models. We then propose a framework built on pre-trained DinoV2 and text-to-image stable diffusion models to infer the normalized object coordinate space (NOCS) maps of the target instances. This framework fully leverages the visual semantic prior from DinoV2 and the aligned visual and language knowledge within the text-to-image diffusion model, which enables generalization to various text descriptions of novel categories. Comprehensive quantitative and qualitative experiments demonstrate that the proposed open-vocabulary method, trained on our large-scale synthesized data, significantly outperforms the baseline and can effectively generalize to real-world images of unseen categories. The project page is at https://ov9d.github.io.

Minglin Chen, Weihao Yuan, Yukun Wang et al.

Recently, text-to-3D approaches have achieved high-fidelity 3D content generation using text description. However, the generated objects are stochastic and lack fine-grained control. Sketches provide a cheap approach to introduce such fine-grained control. Nevertheless, it is challenging to achieve flexible control from these sketches due to their abstraction and ambiguity. In this paper, we present a multi-view sketch-guided text-to-3D generation framework (namely, Sketch2NeRF) to add sketch control to 3D generation. Specifically, our method leverages pretrained 2D diffusion models (e.g., Stable Diffusion and ControlNet) to supervise the optimization of a 3D scene represented by a neural radiance field (NeRF). We propose a novel synchronized generation and reconstruction method to effectively optimize the NeRF. In the experiments, we collected two kinds of multi-view sketch datasets to evaluate the proposed method. We demonstrate that our method can synthesize 3D consistent contents with fine-grained sketch control while being high-fidelity to text prompts. Extensive results show that our method achieves state-of-the-art performance in terms of sketch similarity and text alignment.

Weihao Yuan, Rui Fan, Michael Yu Wang et al.

We design a multiscopic vision system that utilizes a low-cost monocular RGB camera to acquire accurate depth estimation. Unlike multi-view stereo with images captured at unconstrained camera poses, the proposed system controls the motion of a camera to capture a sequence of images in horizontally or vertically aligned positions with the same parallax. In this system, we propose a new heuristic method and a robust learning-based method to fuse multiple cost volumes between the reference image and its surrounding images. To obtain training data, we build a synthetic dataset with multiscopic images. The experiments on the real-world Middlebury dataset and real robot demonstration show that our multiscopic vision system outperforms traditional two-frame stereo matching methods in depth estimation. Our code and dataset are available at https://sites.google.com/view/multiscopic.

Weihao Yuan, Yazhan Zhang, Bingkun Wu et al.

Self-supervised learning for depth estimation possesses several advantages over supervised learning. The benefits of no need for ground-truth depth, online fine-tuning, and better generalization with unlimited data attract researchers to seek self-supervised solutions. In this work, we propose a new self-supervised framework for stereo matching utilizing multiple images captured at aligned camera positions. A cross photometric loss, an uncertainty-aware mutual-supervision loss, and a new smoothness loss are introduced to optimize the network in learning disparity maps end-to-end without ground-truth depth information. To train this framework, we build a new multiscopic dataset consisting of synthetic images rendered by 3D engines and real images captured by real cameras. After being trained with only the synthetic images, our network can perform well in unseen outdoor scenes. Our experiment shows that our model obtains better disparity maps than previous unsupervised methods on the KITTI dataset and is comparable to supervised methods when generalized to unseen data. Our source code and dataset are available at https://sites.google.com/view/multiscopic.

Weihao Yuan, Michael Yu Wang, Qifeng Chen

Self-supervised learning for visual object tracking possesses valuable advantages compared to supervised learning, such as the non-necessity of laborious human annotations and online training. In this work, we exploit an end-to-end Siamese network in a cycle-consistent self-supervised framework for object tracking. Self-supervision can be performed by taking advantage of the cycle consistency in the forward and backward tracking. To better leverage the end-to-end learning of deep networks, we propose to integrate a Siamese region proposal and mask regression network in our tracking framework so that a fast and more accurate tracker can be learned without the annotation of each frame. The experiments on the VOT dataset for visual object tracking and on the DAVIS dataset for video object segmentation propagation show that our method outperforms prior approaches on both tasks.

Weihao Yuan, Rui Fan, Michael Yu Wang et al.

We design a multiscopic vision system that utilizes a low-cost monocular RGB camera to acquire accurate depth estimation for robotic applications. Unlike multi-view stereo with images captured at unconstrained camera poses, the proposed system actively controls a robot arm with a mounted camera to capture a sequence of images in horizontally or vertically aligned positions with the same parallax. In this system, we combine the cost volumes for stereo matching between the reference image and the surrounding images to form a fused cost volume that is robust to outliers. Experiments on the Middlebury dataset and real robot experiments show that our obtained disparity maps are more accurate than two-frame stereo matching: the average absolute error is reduced by 50.2% in our experiments.

Haoran Song, Joshua A. Haustein, Weihao Yuan et al.

In this work, we address a planar non-prehensile sorting task. Here, a robot needs to push many densely packed objects belonging to different classes into a configuration where these classes are clearly separated from each other. To achieve this, we propose to employ Monte Carlo tree search equipped with a task-specific heuristic function. We evaluate the algorithm on various simulated and real-world sorting tasks. We observe that the algorithm is capable to reliably sort large numbers of convex and non-convex objects, as well as convex objects in the presence of immovable obstacles.

Yazhan Zhang, Weihao Yuan, Zicheng Kan et al.

In essence, successful grasp boils down to correct responses to multiple contact events between fingertips and objects. In most scenarios, tactile sensing is adequate to distinguish contact events. Due to the nature of high dimensionality of tactile information, classifying spatiotemporal tactile signals using conventional model-based methods is difficult. In this work, we propose to predict and classify tactile signal using deep learning methods, seeking to enhance the adaptability of the robotic grasp system to external event changes that may lead to grasping failure. We develop a deep learning framework and collect 6650 tactile image sequences with a vision-based tactile sensor, and the neural network is integrated into a contact-event-based robotic grasping system. In grasping experiments, we achieved 52% increase in terms of object lifting success rate with contact detection, significantly higher robustness under unexpected loads with slip prediction compared with open-loop grasps, demonstrating that integration of the proposed framework into robotic grasping system substantially improves picking success rate and capability to withstand external disturbances.

Weihao Yuan, Kaiyu Hang, Haoran Song et al.

Moving a human body or a large and bulky object can require the strength of whole arm manipulation (WAM). This type of manipulation places the load on the robot's arms and relies on global properties of the interaction to succeed---rather than local contacts such as grasping or non-prehensile pushing. In this paper, we learn to generate motions that enable WAM for holding and transporting of humans in certain rescue or patient care scenarios. We model the task as a reinforcement learning problem in order to provide a behavior that can directly respond to external perturbation and human motion. For this, we represent global properties of the robot-human interaction with topology-based coordinates that are computed from arm and torso positions. These coordinates also allow transferring the learned policy to other body shapes and sizes. For training and evaluation, we simulate a dynamic sea rescue scenario and show in quantitative experiments that the policy can solve unseen scenarios with differently-shaped humans, floating humans, or with perception noise. Our qualitative experiments show the subsequent transporting after holding is achieved and we demonstrate that the policy can be directly transferred to a real world setting.

Weihao Yuan, Johannes A. Stork, Danica Kragic et al.

Rearranging objects on a tabletop surface by means of nonprehensile manipulation is a task which requires skillful interaction with the physical world. Usually, this is achieved by precisely modeling physical properties of the objects, robot, and the environment for explicit planning. In contrast, as explicitly modeling the physical environment is not always feasible and involves various uncertainties, we learn a nonprehensile rearrangement strategy with deep reinforcement learning based on only visual feedback. For this, we model the task with rewards and train a deep Q-network. Our potential field-based heuristic exploration strategy reduces the amount of collisions which lead to suboptimal outcomes and we actively balance the training set to avoid bias towards poor examples. Our training process leads to quicker learning and better performance on the task as compared to uniform exploration and standard experience replay. We demonstrate empirical evidence from simulation that our method leads to a success rate of 85%, show that our system can cope with sudden changes of the environment, and compare our performance with human level performance.