Yueqi Duan

Muheng Li, Yueqi Duan, Jie Zhou et al. · tsinghua

With the rising industrial attention to 3D virtual modeling technology, generating novel 3D content based on specified conditions (e.g. text) has become a hot issue. In this paper, we propose a new generative 3D modeling framework called Diffusion-SDF for the challenging task of text-to-shape synthesis. Previous approaches lack flexibility in both 3D data representation and shape generation, thereby failing to generate highly diversified 3D shapes conforming to the given text descriptions. To address this, we propose a SDF autoencoder together with the Voxelized Diffusion model to learn and generate representations for voxelized signed distance fields (SDFs) of 3D shapes. Specifically, we design a novel UinU-Net architecture that implants a local-focused inner network inside the standard U-Net architecture, which enables better reconstruction of patch-independent SDF representations. We extend our approach to further text-to-shape tasks including text-conditioned shape completion and manipulation. Experimental results show that Diffusion-SDF generates both higher quality and more diversified 3D shapes that conform well to given text descriptions when compared to previous approaches. Code is available at: https://github.com/ttlmh/Diffusion-SDF

Shuai Shen, Wanhua Li, Zheng Zhu et al. · tsinghua

Talking head synthesis is an emerging technology with wide applications in film dubbing, virtual avatars and online education. Recent NeRF-based methods generate more natural talking videos, as they better capture the 3D structural information of faces. However, a specific model needs to be trained for each identity with a large dataset. In this paper, we propose Dynamic Facial Radiance Fields (DFRF) for few-shot talking head synthesis, which can rapidly generalize to an unseen identity with few training data. Different from the existing NeRF-based methods which directly encode the 3D geometry and appearance of a specific person into the network, our DFRF conditions face radiance field on 2D appearance images to learn the face prior. Thus the facial radiance field can be flexibly adjusted to the new identity with few reference images. Additionally, for better modeling of the facial deformations, we propose a differentiable face warping module conditioned on audio signals to deform all reference images to the query space. Extensive experiments show that with only tens of seconds of training clip available, our proposed DFRF can synthesize natural and high-quality audio-driven talking head videos for novel identities with only 40k iterations. We highly recommend readers view our supplementary video for intuitive comparisons. Code is available in https://sstzal.github.io/DFRF/.

An Tao, Yueqi Duan, Yingqi Wang et al. · tsinghua

In this paper, we investigate the dynamics-aware adversarial attack problem of adaptive neural networks. Most existing adversarial attack algorithms are designed under a basic assumption -- the network architecture is fixed throughout the attack process. However, this assumption does not hold for many recently proposed adaptive neural networks, which adaptively deactivate unnecessary execution units based on inputs to improve computational efficiency. It results in a serious issue of lagged gradient, making the learned attack at the current step ineffective due to the architecture change afterward. To address this issue, we propose a Leaded Gradient Method (LGM) and show the significant effects of the lagged gradient. More specifically, we reformulate the gradients to be aware of the potential dynamic changes of network architectures, so that the learned attack better "leads" the next step than the dynamics-unaware methods when network architecture changes dynamically. Extensive experiments on representative types of adaptive neural networks for both 2D images and 3D point clouds show that our LGM achieves impressive adversarial attack performance compared with the dynamic-unaware attack methods. Code is available at https://github.com/antao97/LGM.

Wenzhao Zheng, Weiliang Chen, Yuanhui Huang et al. · tsinghua

Understanding how the 3D scene evolves is vital for making decisions in autonomous driving. Most existing methods achieve this by predicting the movements of object boxes, which cannot capture more fine-grained scene information. In this paper, we explore a new framework of learning a world model, OccWorld, in the 3D Occupancy space to simultaneously predict the movement of the ego car and the evolution of the surrounding scenes. We propose to learn a world model based on 3D occupancy rather than 3D bounding boxes and segmentation maps for three reasons: 1) expressiveness. 3D occupancy can describe the more fine-grained 3D structure of the scene; 2) efficiency. 3D occupancy is more economical to obtain (e.g., from sparse LiDAR points). 3) versatility. 3D occupancy can adapt to both vision and LiDAR. To facilitate the modeling of the world evolution, we learn a reconstruction-based scene tokenizer on the 3D occupancy to obtain discrete scene tokens to describe the surrounding scenes. We then adopt a GPT-like spatial-temporal generative transformer to generate subsequent scene and ego tokens to decode the future occupancy and ego trajectory. Extensive experiments on the widely used nuScenes benchmark demonstrate the ability of OccWorld to effectively model the evolution of the driving scenes. OccWorld also produces competitive planning results without using instance and map supervision. Code: https://github.com/wzzheng/OccWorld.

Muheng Li, Lei Chen, Yueqi Duan et al.

Action recognition models have shown a promising capability to classify human actions in short video clips. In a real scenario, multiple correlated human actions commonly occur in particular orders, forming semantically meaningful human activities. Conventional action recognition approaches focus on analyzing single actions. However, they fail to fully reason about the contextual relations between adjacent actions, which provide potential temporal logic for understanding long videos. In this paper, we propose a prompt-based framework, Bridge-Prompt (Br-Prompt), to model the semantics across adjacent actions, so that it simultaneously exploits both out-of-context and contextual information from a series of ordinal actions in instructional videos. More specifically, we reformulate the individual action labels as integrated text prompts for supervision, which bridge the gap between individual action semantics. The generated text prompts are paired with corresponding video clips, and together co-train the text encoder and the video encoder via a contrastive approach. The learned vision encoder has a stronger capability for ordinal-action-related downstream tasks, e.g. action segmentation and human activity recognition. We evaluate the performances of our approach on several video datasets: Georgia Tech Egocentric Activities (GTEA), 50Salads, and the Breakfast dataset. Br-Prompt achieves state-of-the-art on multiple benchmarks. Code is available at https://github.com/ttlmh/Bridge-Prompt

Fengrui Tian, Shaoyi Du, Yueqi Duan

In this paper, we target at the problem of learning a generalizable dynamic radiance field from monocular videos. Different from most existing NeRF methods that are based on multiple views, monocular videos only contain one view at each timestamp, thereby suffering from ambiguity along the view direction in estimating point features and scene flows. Previous studies such as DynNeRF disambiguate point features by positional encoding, which is not transferable and severely limits the generalization ability. As a result, these methods have to train one independent model for each scene and suffer from heavy computational costs when applying to increasing monocular videos in real-world applications. To address this, We propose MonoNeRF to simultaneously learn point features and scene flows with point trajectory and feature correspondence constraints across frames. More specifically, we learn an implicit velocity field to estimate point trajectory from temporal features with Neural ODE, which is followed by a flow-based feature aggregation module to obtain spatial features along the point trajectory. We jointly optimize temporal and spatial features in an end-to-end manner. Experiments show that our MonoNeRF is able to learn from multiple scenes and support new applications such as scene editing, unseen frame synthesis, and fast novel scene adaptation. Codes are available at https://github.com/tianfr/MonoNeRF.

Yu Zheng, Yueqi Duan, Jiwen Lu et al. · tsinghua

A bathtub in a library, a sink in an office, a bed in a laundry room -- the counter-intuition suggests that scene provides important prior knowledge for 3D object detection, which instructs to eliminate the ambiguous detection of similar objects. In this paper, we propose HyperDet3D to explore scene-conditioned prior knowledge for 3D object detection. Existing methods strive for better representation of local elements and their relations without scene-conditioned knowledge, which may cause ambiguity merely based on the understanding of individual points and object candidates. Instead, HyperDet3D simultaneously learns scene-agnostic embeddings and scene-specific knowledge through scene-conditioned hypernetworks. More specifically, our HyperDet3D not only explores the sharable abstracts from various 3D scenes, but also adapts the detector to the given scene at test time. We propose a discriminative Multi-head Scene-specific Attention (MSA) module to dynamically control the layer parameters of the detector conditioned on the fusion of scene-conditioned knowledge. Our HyperDet3D achieves state-of-the-art results on the 3D object detection benchmark of the ScanNet and SUN RGB-D datasets. Moreover, through cross-dataset evaluation, we show the acquired scene-conditioned prior knowledge still takes effect when facing 3D scenes with domain gap.

Chensheng Dai, Shengjun Zhang, Yifan Li et al.

Video generation models based on Diffusion Transformers (DiTs) have achieved remarkable performance in video synthesis, yet they suffer from high inference latency and computational costs due to the quadratic complexity of 3D attention. Existing acceleration methods primarily reduce computational complexity within each individual denoising steps through techniques such as sparse attention and KV-caching. However, they rigidly adhere to the inherent constraint of the standard diffusion pipeline: every frame in the target video sequence must be subjected to a complete, dense denoising process across all diffusion timesteps. We observe that due to the corresponding contents and motions among adjacent frames, when keyframes with critical semantic transitions are anchored, the intermediate states of others often follow more predictable trajectories, which indicates that such uniform, dense denoising process is inherently redundant for natural video data. To this end, we introduce \textbf{RhymeFlow}, a training-free framework that decouples the denoising trajectories of different frames. Specifically, we first identify a sparse set of pivotal key frames that dominate the latent semantic evolution. Then, only these keyframes undergo dense, step-by-step denoising to ensure structural integrity, while non-keyframes progressively skip denoising steps to minimize computational cost. Since skipped intermediate states of non-keyframes break the temporal coherence in keyframe denoising steps, leading to visual degradation, we further introduce a latent trajectory projection module, which enables keyframes to interact with a complete and temporally consistent sequence representation. Extensive experiments on current DiT-based video generation models demonstrate our method outperforms existing baselines with higher inference speed and better visual quality.

Yichen Li, Kaichun Mo, Yueqi Duan et al.

Shape assembly composes complex shapes geometries by arranging simple part geometries and has wide applications in autonomous robotic assembly and CAD modeling. Existing works focus on geometry reasoning and neglect the actual physical assembly process of matching and fitting joints, which are the contact surfaces connecting different parts. In this paper, we consider contacting joints for the task of multi-part assembly. A successful joint-optimized assembly needs to satisfy the bilateral objectives of shape structure and joint alignment. We propose a hierarchical graph learning approach composed of two levels of graph representation learning. The part graph takes part geometries as input to build the desired shape structure. The joint-level graph uses part joints information and focuses on matching and aligning joints. The two kinds of information are combined to achieve the bilateral objectives. Extensive experiments demonstrate that our method outperforms previous methods, achieving better shape structure and higher joint alignment accuracy.

Runzhong Zhang, Suchen Wang, Yueqi Duan et al.

In this paper, we propose an HOI-aware adaptive network named AdaAct for weakly-supervised action segmentation. Most existing methods learn a fixed network to predict the action of each frame with the neighboring frames. However, this would result in ambiguity when estimating similar actions, such as pouring juice and pouring coffee. To address this, we aim to exploit temporally global but spatially local human-object interactions (HOI) as video-level prior knowledge for action segmentation. The long-term HOI sequence provides crucial contextual information to distinguish ambiguous actions, where our network dynamically adapts to the given HOI sequence at test time. More specifically, we first design a video HOI encoder that extracts, selects, and integrates the most representative HOI throughout the video. Then, we propose a two-branch HyperNetwork to learn an adaptive temporal encoder, which automatically adjusts the parameters based on the HOI information of various videos on the fly. Extensive experiments on two widely-used datasets including Breakfast and 50Salads demonstrate the effectiveness of our method under different evaluation metrics.

Fangfu Liu, Chubin Zhang, Yu Zheng et al.

In this paper, we aim to learn a semantic radiance field from multiple scenes that is accurate, efficient and generalizable. While most existing NeRFs target at the tasks of neural scene rendering, image synthesis and multi-view reconstruction, there are a few attempts such as Semantic-NeRF that explore to learn high-level semantic understanding with the NeRF structure. However, Semantic-NeRF simultaneously learns color and semantic label from a single ray with multiple heads, where the single ray fails to provide rich semantic information. As a result, Semantic NeRF relies on positional encoding and needs to train one specific model for each scene. To address this, we propose Semantic Ray (S-Ray) to fully exploit semantic information along the ray direction from its multi-view reprojections. As directly performing dense attention over multi-view reprojected rays would suffer from heavy computational cost, we design a Cross-Reprojection Attention module with consecutive intra-view radial and cross-view sparse attentions, which decomposes contextual information along reprojected rays and cross multiple views and then collects dense connections by stacking the modules. Experiments show that our S-Ray is able to learn from multiple scenes, and it presents strong generalization ability to adapt to unseen scenes.

Xiaoyu Feng, Heming Du, Yueqi Duan et al.

Effectively preserving and encoding structure features from objects in irregular and sparse LiDAR points is a key challenge to 3D object detection on point cloud. Recently, Transformer has demonstrated promising performance on many 2D and even 3D vision tasks. Compared with the fixed and rigid convolution kernels, the self-attention mechanism in Transformer can adaptively exclude the unrelated or noisy points and thus suitable for preserving the local spatial structure in irregular LiDAR point cloud. However, Transformer only performs a simple sum on the point features, based on the self-attention mechanism, and all the points share the same transformation for value. Such isotropic operation lacks the ability to capture the direction-distance-oriented local structure which is important for 3D object detection. In this work, we propose a Structure-Embedding transFormer (SEFormer), which can not only preserve local structure as traditional Transformer but also have the ability to encode the local structure. Compared to the self-attention mechanism in traditional Transformer, SEFormer learns different feature transformations for value points based on the relative directions and distances to the query point. Then we propose a SEFormer based network for high-performance 3D object detection. Extensive experiments show that the proposed architecture can achieve SOTA results on Waymo Open Dataset, the largest 3D detection benchmark for autonomous driving. Specifically, SEFormer achieves 79.02% mAP, which is 1.2% higher than existing works. We will release the codes.

Shengjun Zhang, Zhang Zhang, Simin Huang et al.

Recent advancements in video-based world models have demonstrated an unprecedented ability to synthesize high-fidelity visual sequences. However, a fundamental gap persists between visually plausible video generation and the functional requirements of a world model, particularly in maintaining a stable and reasonable internal state over extended temporal horizons. While existing benchmarks primarily emphasize visual quality, motion coherence, and text-video alignment, they largely overlook memory, the core capability of a world model to preserve consistency across long-term horizons and complex interactions. To address this gap, we present \textbf{MBench}, a comprehensive benchmark dedicated to quantifying and evaluating the memory capability of video world models. We systematically decompose the memory capability of video world models into three hierarchical and complementary core dimensions: entity consistency, environment consistency, and causal consistency, which are further refined into 12 quantifiable sub-dimensions for comprehensive characterization of long-term memory. Our benchmark is built upon rigorously curated real-captured long videos, and evaluated by rule-based quantitative matrices and VLM to enable objective and comprehensive consistency assessment. Extensive evaluations of mainstream state-of-the-art video world models reveal critical systemic limitations of existing methods in long-term state retention, providing a standardized benchmark and clear research direction to advance the field.

Fangfu Liu, Wenchang Ma, An Zhang et al.

Discovering causal structure from purely observational data (i.e., causal discovery), aiming to identify causal relationships among variables, is a fundamental task in machine learning. The recent invention of differentiable score-based DAG learners is a crucial enabler, which reframes the combinatorial optimization problem into a differentiable optimization with a DAG constraint over directed graph space. Despite their great success, these cutting-edge DAG learners incorporate DAG-ness independent score functions to evaluate the directed graph candidates, lacking in considering graph structure. As a result, measuring the data fitness alone regardless of DAG-ness inevitably leads to discovering suboptimal DAGs and model vulnerabilities. Towards this end, we propose a dynamic causal space for DAG structure learning, coined CASPER, that integrates the graph structure into the score function as a new measure in the causal space to faithfully reflect the causal distance between estimated and ground truth DAG. CASPER revises the learning process as well as enhances the DAG structure learning via adaptive attention to DAG-ness. Grounded by empirical visualization, CASPER, as a space, satisfies a series of desired properties, such as structure awareness and noise robustness. Extensive experiments on both synthetic and real-world datasets clearly validate the superiority of our CASPER over the state-of-the-art causal discovery methods in terms of accuracy and robustness.

Yang Zheng, Tolga Birdal, Fei Xia et al.

We propose a novel method to reliably estimate the pose of a camera given a sequence of images acquired in extreme environments such as deep seas or extraterrestrial terrains. Data acquired under these challenging conditions are corrupted by textureless surfaces, image degradation, and presence of repetitive and highly ambiguous structures. When naively deployed, the state-of-the-art methods can fail in those scenarios as confirmed by our empirical analysis. In this paper, we attempt to make camera relocalization work in these extreme situations. To this end, we propose: (i) a hierarchical localization system, where we leverage temporal information and (ii) a novel environment-aware image enhancement method to boost the robustness and accuracy. Our extensive experimental results demonstrate superior performance in favor of our method under two extreme settings: localizing an autonomous underwater vehicle and localizing a planetary rover in a Mars-like desert. In addition, our method achieves comparable performance with state-of-the-art methods on the indoor benchmark (7-Scenes dataset) using only 20% training data.

Fangfu Liu, Wenqiang Sun, Hanyang Wang et al.

Advancements in 3D scene reconstruction have transformed 2D images from the real world into 3D models, producing realistic 3D results from hundreds of input photos. Despite great success in dense-view reconstruction scenarios, rendering a detailed scene from insufficient captured views is still an ill-posed optimization problem, often resulting in artifacts and distortions in unseen areas. In this paper, we propose ReconX, a novel 3D scene reconstruction paradigm that reframes the ambiguous reconstruction challenge as a temporal generation task. The key insight is to unleash the strong generative prior of large pre-trained video diffusion models for sparse-view reconstruction. However, 3D view consistency struggles to be accurately preserved in directly generated video frames from pre-trained models. To address this, given limited input views, the proposed ReconX first constructs a global point cloud and encodes it into a contextual space as the 3D structure condition. Guided by the condition, the video diffusion model then synthesizes video frames that are both detail-preserved and exhibit a high degree of 3D consistency, ensuring the coherence of the scene from various perspectives. Finally, we recover the 3D scene from the generated video through a confidence-aware 3D Gaussian Splatting optimization scheme. Extensive experiments on various real-world datasets show the superiority of our ReconX over state-of-the-art methods in terms of quality and generalizability.

Fangfu Liu, Kai He, Tianchang Shen et al.

World models for interactive video generation have largely focused on single-agent settings, where future observations are generated from a single control signal. However, many generated environments require multi-agent interaction: multiple players, robots, or embodied agents act simultaneously within a shared space. Scaling world models to such settings requires a principled multi-agent design: agents should remain independently controllable, permutation-symmetric, and support efficient inference while maintaining consistency across time and perspectives. In this paper, we present our generative multi-agent world model for interactive simulation. It introduces Simplex Rotary Agent Encoding, a parameter-free extension of 3D RoPE that represents agents as vertices of a regular simplex in rotary angle space. This gives each agent a distinct phase while making all agents permutation-equivalent, enabling scalable agent identity without learned per-slot identities or a fixed agent ordering. To avoid dense all-to-all attention across agents, we further propose Sparse Hub Attention, where learnable hub tokens mediate token interaction across agents, reducing cross-agent attention cost from quadratic to linear in the number of agents. For real-time rollout, we distill a full-context diffusion teacher into a causal student that generates temporal blocks sequentially with KV caching, enabling action-responsive generation at 24 FPS. Experiments in multiplayer virtual environments show that our model improves video fidelity, action controllability, and inter-agent consistency over slot-based and dense-attention baselines, while generalizing from two to four players without additional training.

Yu Zheng, Yueqi Duan, Kangfu Zheng et al.

In this paper, we propose a One-Point-One NeRF (OPONeRF) framework for robust scene rendering. Existing NeRFs are designed based on a key assumption that the target scene remains unchanged between the training and test time. However, small but unpredictable perturbations such as object movements, light changes and data contaminations broadly exist in real-life 3D scenes, which lead to significantly defective or failed rendering results even for the recent state-of-the-art generalizable methods. To address this, we propose a divide-and-conquer framework in OPONeRF that adaptively responds to local scene variations via personalizing appropriate point-wise parameters, instead of fitting a single set of NeRF parameters that are inactive to test-time unseen changes. Moreover, to explicitly capture the local uncertainty, we decompose the point representation into deterministic mapping and probabilistic inference. In this way, OPONeRF learns the sharable invariance and unsupervisedly models the unexpected scene variations between the training and testing scenes. To validate the effectiveness of the proposed method, we construct benchmarks from both realistic and synthetic data with diverse test-time perturbations including foreground motions, illumination variations and multi-modality noises, which are more challenging than conventional generalization and temporal reconstruction benchmarks. Experimental results show that our OPONeRF outperforms state-of-the-art NeRFs on various evaluation metrics through benchmark experiments and cross-scene evaluations. We further show the efficacy of the proposed method via experimenting on other existing generalization-based benchmarks and incorporating the idea of One-Point-One NeRF into other advanced baseline methods.

Weiliang Chen, Fangfu Liu, Diankun Wu et al.

We are living in a flourishing era of digital media, where everyone has the potential to become a personal filmmaker. Current research on video generation suggests a promising avenue for controllable film creation in pixel space using Diffusion models. However, the reliance on overly verbose prompts and insufficient focus on cinematic elements (e.g., camera movement) results in videos that lack cinematic quality. Furthermore, the absence of 3D modeling often leads to failures in video generation, such as inconsistent character models at different frames, ultimately hindering the immersive experience for viewers. In this paper, we propose a new framework for film creation, Dream-Cinema, which is designed for user-friendly, 3D space-based film creation with generative models. Specifically, we decompose 3D film creation into four key elements: 3D character, driven motion, camera movement, and environment. We extract the latter three elements from user-specified film shots and generate the 3D character using a generative model based on a provided image. To seamlessly recombine these elements and ensure smooth film creation, we propose structure-guided character animation, shape-aware camera movement optimization, and environment-aware generative refinement. Extensive experiments demonstrate the effectiveness of our method in generating high-quality films with free camera and 3D characters.

Hanyang Wang, Yiyang Liu, Jiawei Chi et al.

Classifier-Free Guidance (CFG) has emerged as a central approach for enhancing semantic alignment in flow-based diffusion models. In this paper, we explore a unified framework called CFG-Ctrl, which reinterprets CFG as a control applied to the first-order continuous-time generative flow, using the conditional-unconditional discrepancy as an error signal to adjust the velocity field. From this perspective, we summarize vanilla CFG as a proportional controller (P-control) with fixed gain, and typical follow-up variants develop extended control-law designs derived from it. However, existing methods mainly rely on linear control, inherently leading to instability, overshooting, and degraded semantic fidelity especially on large guidance scales. To address this, we introduce Sliding Mode Control CFG (SMC-CFG), which enforces the generative flow toward a rapidly convergent sliding manifold. Specifically, we define an exponential sliding mode surface over the semantic prediction error and introduce a switching control term to establish nonlinear feedback-guided correction. Moreover, we provide a Lyapunov stability analysis to theoretically support finite-time convergence. Experiments across text-to-image generation models including Stable Diffusion 3.5, Flux, and Qwen-Image demonstrate that SMC-CFG outperforms standard CFG in semantic alignment and enhances robustness across a wide range of guidance scales. Project Page: https://hanyang-21.github.io/CFG-Ctrl

Chong Xia, Kai Zhu, Zizhuo Wang et al.

Compositional scene reconstruction seeks to create object-centric representations rather than holistic scenes from real-world videos, which is natively applicable for simulation and interaction. Conventional compositional reconstruction approaches primarily emphasize on visual appearance and show limited generalization ability to real-world scenarios. In this paper, we propose SimRecon, a framework that realizes a "Perception-Generation-Simulation" pipeline towards cluttered scene reconstruction, which first conducts scene-level semantic reconstruction from video input, then performs single-object generation, and finally assembles these assets in the simulator. However, naively combining these three stages leads to visual infidelity of generated assets and physical implausibility of the final scene, a problem particularly severe for complex scenes. Thus, we further propose two bridging modules between the three stages to address this problem. To be specific, for the transition from Perception to Generation, critical for visual fidelity, we introduce Active Viewpoint Optimization, which actively searches in 3D space to acquire optimal projected images as conditions for single-object completion. Moreover, for the transition from Generation to Simulation, essential for physical plausibility, we propose a Scene Graph Synthesizer, which guides the construction from scratch in 3D simulators, mirroring the native, constructive principle of the real world. Extensive experiments on the ScanNet dataset validate our method's superior performance over previous state-of-the-art approaches.

Xin Fei, Wenzhao Zheng, Yueqi Duan et al.

We propose PixelGaussian, an efficient feed-forward framework for learning generalizable 3D Gaussian reconstruction from arbitrary views. Most existing methods rely on uniform pixel-wise Gaussian representations, which learn a fixed number of 3D Gaussians for each view and cannot generalize well to more input views. Differently, our PixelGaussian dynamically adapts both the Gaussian distribution and quantity based on geometric complexity, leading to more efficient representations and significant improvements in reconstruction quality. Specifically, we introduce a Cascade Gaussian Adapter to adjust Gaussian distribution according to local geometry complexity identified by a keypoint scorer. CGA leverages deformable attention in context-aware hypernetworks to guide Gaussian pruning and splitting, ensuring accurate representation in complex regions while reducing redundancy. Furthermore, we design a transformer-based Iterative Gaussian Refiner module that refines Gaussian representations through direct image-Gaussian interactions. Our PixelGaussian can effectively reduce Gaussian redundancy as input views increase. We conduct extensive experiments on the large-scale ACID and RealEstate10K datasets, where our method achieves state-of-the-art performance with good generalization to various numbers of views. Code: https://github.com/Barrybarry-Smith/PixelGaussian.

Xin Fei, Wenzhao Zheng, Yueqi Duan et al.

Realtime 4D reconstruction for dynamic scenes remains a crucial challenge for autonomous driving perception. Most existing methods rely on depth estimation through self-supervision or multi-modality sensor fusion. In this paper, we propose Driv3R, a DUSt3R-based framework that directly regresses per-frame point maps from multi-view image sequences. To achieve streaming dense reconstruction, we maintain a memory pool to reason both spatial relationships across sensors and dynamic temporal contexts to enhance multi-view 3D consistency and temporal integration. Furthermore, we employ a 4D flow predictor to identify moving objects within the scene to direct our network focus more on reconstructing these dynamic regions. Finally, we align all per-frame pointmaps consistently to the world coordinate system in an optimization-free manner. We conduct extensive experiments on the large-scale nuScenes dataset to evaluate the effectiveness of our method. Driv3R outperforms previous frameworks in 4D dynamic scene reconstruction, achieving 15x faster inference speed compared to methods requiring global alignment. Code: https://github.com/Barrybarry-Smith/Driv3R.

Fangfu Liu, Diankun Wu, Jiawei Chi et al.

Humans perceive and understand real-world spaces through a stream of visual observations. Therefore, the ability to streamingly maintain and update spatial evidence from potentially unbounded video streams is essential for spatial intelligence. The core challenge is not simply longer context windows but how spatial information is selected, organized, and retained over time. In this paper, we propose Spatial-TTT towards streaming visual-based spatial intelligence with test-time training (TTT), which adapts a subset of parameters (fast weights) to capture and organize spatial evidence over long-horizon scene videos. Specifically, we design a hybrid architecture and adopt large-chunk updates parallel with sliding-window attention for efficient spatial video processing. To further promote spatial awareness, we introduce a spatial-predictive mechanism applied to TTT layers with 3D spatiotemporal convolution, which encourages the model to capture geometric correspondence and temporal continuity across frames. Beyond architecture design, we construct a dataset with dense 3D spatial descriptions, which guides the model to update its fast weights to memorize and organize global 3D spatial signals in a structured manner. Extensive experiments demonstrate that Spatial-TTT improves long-horizon spatial understanding and achieves state-of-the-art performance on video spatial benchmarks. Project page: https://liuff19.github.io/Spatial-TTT.

Chong Xia, Fangfu Liu, Yule Wang et al.

Recent advances in generalizable 3D Gaussian Splatting (3DGS) have enabled rapid 3D scene reconstruction within seconds, eliminating the need for per-scene optimization. However, existing methods primarily follow an offline reconstruction paradigm, lacking the capacity for continuous reconstruction, which limits their applicability to online scenarios such as robotics and VR/AR. In this paper, we introduce OnlineX, a feed-forward framework that reconstructs both 3D visual appearance and language fields in an online manner using only streaming images. A key challenge in online formulation is the cumulative drift issue, which is rooted in the fundamental conflict between two opposing roles of the memory state: an active role that constantly refreshes to capture high-frequency local geometry, and a stable role that conservatively accumulates and preserves the long-term global structure. To address this, we introduce a decoupled active-to-stable state evolution paradigm. Our framework decouples the memory state into a dedicated active state and a persistent stable state, and then cohesively fuses the information from the former into the latter to achieve both fidelity and stability. Moreover, we jointly model visual appearance and language fields and incorporate an implicit Gaussian fusion module to enhance reconstruction quality. Experiments on mainstream datasets demonstrate that our method consistently outperforms prior work in novel view synthesis and semantic understanding, showcasing robust performance across input sequences of varying lengths with real-time inference speed.

Fengrui Tian, Yueqi Duan, Angtian Wang et al.

In this work, we pioneer Semantic Flow, a neural semantic representation of dynamic scenes from monocular videos. In contrast to previous NeRF methods that reconstruct dynamic scenes from the colors and volume densities of individual points, Semantic Flow learns semantics from continuous flows that contain rich 3D motion information. As there is 2D-to-3D ambiguity problem in the viewing direction when extracting 3D flow features from 2D video frames, we consider the volume densities as opacity priors that describe the contributions of flow features to the semantics on the frames. More specifically, we first learn a flow network to predict flows in the dynamic scene, and propose a flow feature aggregation module to extract flow features from video frames. Then, we propose a flow attention module to extract motion information from flow features, which is followed by a semantic network to output semantic logits of flows. We integrate the logits with volume densities in the viewing direction to supervise the flow features with semantic labels on video frames. Experimental results show that our model is able to learn from multiple dynamic scenes and supports a series of new tasks such as instance-level scene editing, semantic completions, dynamic scene tracking and semantic adaption on novel scenes. Codes are available at https://github.com/tianfr/Semantic-Flow/.

Yang Chen, Yueqi Duan, Haowen Sun et al.

This paper proposes an adaptive margin contrastive learning method for 3D semantic segmentation on point clouds. Most existing methods use equally penalized objectives, which ignore the per-point ambiguities and less discriminated features stemming from transition regions. However, as highly ambiguous points may be indistinguishable even for humans, their manually annotated labels are less reliable, and hard constraints over these points would lead to sub-optimal models. To address this, we first design AMContrast3D, a method comprising contrastive learning into an ambiguity estimation framework, tailored to adaptive objectives for individual points based on ambiguity levels. As a result, our method promotes model training, which ensures the correctness of low-ambiguity points while allowing mistakes for high-ambiguity points. As ambiguities are formulated based on position discrepancies across labels, optimization during inference is constrained by the assumption that all unlabeled points are uniformly unambiguous, lacking ambiguity awareness. Inspired by the insight of joint training, we further propose AMContrast3D++ integrating with two branches trained in parallel, where a novel ambiguity prediction module concurrently learns point ambiguities from generated embeddings. To this end, we design a masked refinement mechanism that leverages predicted ambiguities to enable the ambiguous embeddings to be more reliable, thereby boosting segmentation performance and enhancing robustness. Experimental results on 3D indoor scene datasets, S3DIS and ScanNet, demonstrate the effectiveness of the proposed method. Code is available at https://github.com/YangChenApril/AMContrast3D.

Shengjun Zhang, Zhang Zhang, Chensheng Dai et al.

Recent reinforcement learning has enhanced the flow matching models on human preference alignment. While stochastic sampling enables the exploration of denoising directions, existing methods which optimize over multiple denoising steps suffer from sparse and ambiguous reward signals. We observe that the high entropy steps enable more efficient and effective exploration while the low entropy steps result in undistinguished roll-outs. To this end, we propose E-GRPO, an entropy aware Group Relative Policy Optimization to increase the entropy of SDE sampling steps. Since the integration of stochastic differential equations suffer from ambiguous reward signals due to stochasticity from multiple steps, we specifically merge consecutive low entropy steps to formulate one high entropy step for SDE sampling, while applying ODE sampling on other steps. Building upon this, we introduce multi-step group normalized advantage, which computes group-relative advantages within samples sharing the same consolidated SDE denoising step. Experimental results on different reward settings have demonstrated the effectiveness of our methods.

Mingyu Dong, Chong Xia, Mingyuan Jia et al.

Humans exhibit an innate capacity to rapidly perceive and segment objects from video observations, and even mentally assemble them into structured 3D scenes. Replicating such capability, termed compositional 3D reconstruction, is pivotal for the advancement of Spatial Intelligence and Embodied AI. However, existing methods struggle to achieve practical deployment due to the insufficient integration of cross-modal information, leaving them dependent on manual object prompting, reliant on auxiliary visual inputs, and restricted to overly simplistic scenes by training biases. To address these limitations, we propose ReplicateAnyScene, a framework capable of fully automated and zero-shot transformation of casually captured videos into compositional 3D scenes. Specifically, our pipeline incorporates a five-stage cascade to extract and structurally align generic priors from vision foundation models across textual, visual, and spatial dimensions, grounding them into structured 3D representations and ensuring semantic coherence and physical plausibility of the constructed scenes. To facilitate a more comprehensive evaluation of this task, we further introduce the C3DR benchmark to assess reconstruction quality from diverse aspects. Extensive experiments demonstrate the superiority of our method over existing baselines in generating high-quality compositional 3D scenes.

Yanran Zhang, Bingyao Yu, Yu Zheng et al. · tsinghua

The emergence of visual autoregressive (AR) models has revolutionized image generation while presenting new challenges for synthetic image detection. Unlike previous GAN or diffusion-based methods, AR models generate images through discrete token prediction, exhibiting both marked improvements in image synthesis quality and unique characteristics in their vector-quantized representations. In this paper, we propose to leverage Discrete Distribution Discrepancy-aware Quantization Error (D$^3$QE) for autoregressive-generated image detection that exploits the distinctive patterns and the frequency distribution bias of the codebook existing in real and fake images. We introduce a discrete distribution discrepancy-aware transformer that integrates dynamic codebook frequency statistics into its attention mechanism, fusing semantic features and quantization error latent. To evaluate our method, we construct a comprehensive dataset termed ARForensics covering 7 mainstream visual AR models. Experiments demonstrate superior detection accuracy and strong generalization of D$^3$QE across different AR models, with robustness to real-world perturbations. Code is available at \href{https://github.com/Zhangyr2022/D3QE}{https://github.com/Zhangyr2022/D3QE}.

Yu Zheng, Boyang Gong, Fanye Kong et al.

In this paper, we propose a Counterfactually Decoupled Attention Learning (CDAL) method for open-world model attribution. Existing methods rely on handcrafted design of region partitioning or feature space, which could be confounded by the spurious statistical correlations and struggle with novel attacks in open-world scenarios. To address this, CDAL explicitly models the causal relationships between the attentional visual traces and source model attribution, and counterfactually decouples the discriminative model-specific artifacts from confounding source biases for comparison. In this way, the resulting causal effect provides a quantification on the quality of learned attention maps, thus encouraging the network to capture essential generation patterns that generalize to unseen source models by maximizing the effect. Extensive experiments on existing open-world model attribution benchmarks show that with minimal computational overhead, our method consistently improves state-of-the-art models by large margins, particularly for unseen novel attacks. Source code: https://github.com/yzheng97/CDAL.

Chuanyu Pan, Yanchao Yang, Kaichun Mo et al.

We propose a framework to continuously learn object-centric representations for visual learning and understanding. Existing object-centric representations either rely on supervisions that individualize objects in the scene, or perform unsupervised disentanglement that can hardly deal with complex scenes in the real world. To mitigate the annotation burden and relax the constraints on the statistical complexity of the data, our method leverages interactions to effectively sample diverse variations of an object and the corresponding training signals while learning the object-centric representations. Throughout learning, objects are streamed one by one in random order with unknown identities, and are associated with latent codes that can synthesize discriminative weights for each object through a convolutional hypernetwork. Moreover, re-identification of learned objects and forgetting prevention are employed to make the learning process efficient and robust. We perform an extensive study of the key features of the proposed framework and analyze the characteristics of the learned representations. Furthermore, we demonstrate the capability of the proposed framework in learning representations that can improve label efficiency in downstream tasks. Our code and trained models are made publicly available at: https://github.com/pptrick/Object-Pursuit.

An Tao, Yueqi Duan, Yi Wei et al.

Most existing point cloud instance and semantic segmentation methods rely heavily on strong supervision signals, which require point-level labels for every point in the scene. However, such strong supervision suffers from large annotation costs, arousing the need to study efficient annotating. In this paper, we discover that the locations of instances matter for both instance and semantic 3D scene segmentation. By fully taking advantage of locations, we design a weakly-supervised point cloud segmentation method that only requires clicking on one point per instance to indicate its location for annotation. With over-segmentation for pre-processing, we extend these location annotations into segments as seg-level labels. We further design a segment grouping network (SegGroup) to generate point-level pseudo labels under seg-level labels by hierarchically grouping the unlabeled segments into the relevant nearby labeled segments, so that existing point-level supervised segmentation models can directly consume these pseudo labels for training. Experimental results show that our seg-level supervised method (SegGroup) achieves comparable results with the fully annotated point-level supervised methods. Moreover, it outperforms the recent weakly-supervised methods given a fixed annotation budget. Code is available at https://github.com/AnTao97/SegGroup.

Ziyi Wu, Yueqi Duan, He Wang et al.

Point cloud is an important 3D data representation widely used in many essential applications. Leveraging deep neural networks, recent works have shown great success in processing 3D point clouds. However, those deep neural networks are vulnerable to various 3D adversarial attacks, which can be summarized as two primary types: point perturbation that affects local point distribution, and surface distortion that causes dramatic changes in geometry. In this paper, we simultaneously address both the aforementioned attacks by learning to restore the clean point clouds from the attacked ones. More specifically, we propose an IF-Defense framework to directly optimize the coordinates of input points with geometry-aware and distribution-aware constraints. The former aims to recover the surface of point cloud through implicit function, while the latter encourages evenly-distributed points. Our experimental results show that IF-Defense achieves the state-of-the-art defense performance against existing 3D adversarial attacks on PointNet, PointNet++, DGCNN, PointConv and RS-CNN. For example, compared with previous methods, IF-Defense presents 20.02% improvement in classification accuracy against salient point dropping attack and 16.29% against LG-GAN attack on PointNet. Our code is available at https://github.com/Wuziyi616/IF-Defense.

Wenqiang Sun, Shuo Chen, Fangfu Liu et al.

In this paper, we introduce \textbf{DimensionX}, a framework designed to generate photorealistic 3D and 4D scenes from just a single image with video diffusion. Our approach begins with the insight that both the spatial structure of a 3D scene and the temporal evolution of a 4D scene can be effectively represented through sequences of video frames. While recent video diffusion models have shown remarkable success in producing vivid visuals, they face limitations in directly recovering 3D/4D scenes due to limited spatial and temporal controllability during generation. To overcome this, we propose ST-Director, which decouples spatial and temporal factors in video diffusion by learning dimension-aware LoRAs from dimension-variant data. This controllable video diffusion approach enables precise manipulation of spatial structure and temporal dynamics, allowing us to reconstruct both 3D and 4D representations from sequential frames with the combination of spatial and temporal dimensions. Additionally, to bridge the gap between generated videos and real-world scenes, we introduce a trajectory-aware mechanism for 3D generation and an identity-preserving denoising strategy for 4D generation. Extensive experiments on various real-world and synthetic datasets demonstrate that DimensionX achieves superior results in controllable video generation, as well as in 3D and 4D scene generation, compared with previous methods.

Junliang Ye, Fangfu Liu, Qixiu Li et al.

3D content creation from text prompts has shown remarkable success recently. However, current text-to-3D methods often generate 3D results that do not align well with human preferences. In this paper, we present a comprehensive framework, coined DreamReward, to learn and improve text-to-3D models from human preference feedback. To begin with, we collect 25k expert comparisons based on a systematic annotation pipeline including rating and ranking. Then, we build Reward3D -- the first general-purpose text-to-3D human preference reward model to effectively encode human preferences. Building upon the 3D reward model, we finally perform theoretical analysis and present the Reward3D Feedback Learning (DreamFL), a direct tuning algorithm to optimize the multi-view diffusion models with a redefined scorer. Grounded by theoretical proof and extensive experiment comparisons, our DreamReward successfully generates high-fidelity and 3D consistent results with significant boosts in prompt alignment with human intention. Our results demonstrate the great potential for learning from human feedback to improve text-to-3D models.

Diankun Wu, Fangfu Liu, Yi-Hsin Hung et al.

Recent advancements in Multimodal Large Language Models (MLLMs) have significantly enhanced performance on 2D visual tasks. However, improving their spatial intelligence remains a challenge. Existing 3D MLLMs always rely on additional 3D or 2.5D data to incorporate spatial awareness, restricting their utility in scenarios with only 2D inputs, such as images or videos. In this paper, we present Spatial-MLLM, a novel framework for visual-based spatial reasoning from purely 2D observations. Unlike conventional video MLLMs which rely on CLIP-based visual encoders optimized for semantic understanding, our key insight is to unleash the strong structure prior from the feed-forward visual geometry foundation model. Specifically, we propose a dual-encoder architecture: a pretrained 2D visual encoder to extract semantic features, and a spatial encoder-initialized from the backbone of the visual geometry model-to extract 3D structure features. A connector then integrates both features into unified visual tokens for enhanced spatial understanding. Furthermore, we propose a space-aware frame sampling strategy at inference time, which selects the spatially informative frames of a video sequence, ensuring that even under limited token length, the model focuses on frames critical for spatial reasoning. Beyond architecture improvements, we construct the Spatial-MLLM-120k dataset and train the model on it using supervised fine-tuning and GRPO. Extensive experiments on various real-world datasets demonstrate that our spatial-MLLM achieves state-of-the-art performance in a wide range of visual-based spatial understanding and reasoning tasks. Project page: https://diankun-wu.github.io/Spatial-MLLM/.

Fangfu Liu, Diankun Wu, Yi Wei et al.

Recently, 3D content creation from text prompts has demonstrated remarkable progress by utilizing 2D and 3D diffusion models. While 3D diffusion models ensure great multi-view consistency, their ability to generate high-quality and diverse 3D assets is hindered by the limited 3D data. In contrast, 2D diffusion models find a distillation approach that achieves excellent generalization and rich details without any 3D data. However, 2D lifting methods suffer from inherent view-agnostic ambiguity thereby leading to serious multi-face Janus issues, where text prompts fail to provide sufficient guidance to learn coherent 3D results. Instead of retraining a costly viewpoint-aware model, we study how to fully exploit easily accessible coarse 3D knowledge to enhance the prompts and guide 2D lifting optimization for refinement. In this paper, we propose Sherpa3D, a new text-to-3D framework that achieves high-fidelity, generalizability, and geometric consistency simultaneously. Specifically, we design a pair of guiding strategies derived from the coarse 3D prior generated by the 3D diffusion model: a structural guidance for geometric fidelity and a semantic guidance for 3D coherence. Employing the two types of guidance, the 2D diffusion model enriches the 3D content with diversified and high-quality results. Extensive experiments show the superiority of our Sherpa3D over the state-of-the-art text-to-3D methods in terms of quality and 3D consistency.

Chubin Zhang, Juncheng Yan, Yi Wei et al.

Occupancy prediction reconstructs 3D structures of surrounding environments. It provides detailed information for autonomous driving planning and navigation. However, most existing methods heavily rely on the LiDAR point clouds to generate occupancy ground truth, which is not available in the vision-based system. In this paper, we propose an OccNeRF method for training occupancy networks without 3D supervision. Different from previous works which consider a bounded scene, we parameterize the reconstructed occupancy fields and reorganize the sampling strategy to align with the cameras' infinite perceptive range. The neural rendering is adopted to convert occupancy fields to multi-camera depth maps, supervised by multi-frame photometric consistency. Moreover, for semantic occupancy prediction, we design several strategies to polish the prompts and filter the outputs of a pretrained open-vocabulary 2D segmentation model. Extensive experiments for both self-supervised depth estimation and 3D occupancy prediction tasks on nuScenes and SemanticKITTI datasets demonstrate the effectiveness of our method.

Fangfu Liu, Hanyang Wang, Yimo Cai et al.

With the scale capability of increasing training data, model size, and computational cost, video generation has achieved impressive results in digital creation, enabling users to express creativity across various domains. Recently, researchers in Large Language Models (LLMs) have expanded the scaling to test-time, which can significantly improve LLM performance by using more inference-time computation. Instead of scaling up video foundation models through expensive training costs, we explore the power of Test-Time Scaling (TTS) in video generation, aiming to answer the question: if a video generation model is allowed to use non-trivial amount of inference-time compute, how much can it improve generation quality given a challenging text prompt. In this work, we reinterpret the test-time scaling of video generation as a searching problem to sample better trajectories from Gaussian noise space to the target video distribution. Specifically, we build the search space with test-time verifiers to provide feedback and heuristic algorithms to guide searching process. Given a text prompt, we first explore an intuitive linear search strategy by increasing noise candidates at inference time. As full-step denoising all frames simultaneously requires heavy test-time computation costs, we further design a more efficient TTS method for video generation called Tree-of-Frames (ToF) that adaptively expands and prunes video branches in an autoregressive manner. Extensive experiments on text-conditioned video generation benchmarks demonstrate that increasing test-time compute consistently leads to significant improvements in the quality of videos. Project page: https://liuff19.github.io/Video-T1

Xiuwei Xu, Chong Xia, Ziwei Wang et al.

In this paper, we propose a new framework for online 3D scene perception. Conventional 3D scene perception methods are offline, i.e., take an already reconstructed 3D scene geometry as input, which is not applicable in robotic applications where the input data is streaming RGB-D videos rather than a complete 3D scene reconstructed from pre-collected RGB-D videos. To deal with online 3D scene perception tasks where data collection and perception should be performed simultaneously, the model should be able to process 3D scenes frame by frame and make use of the temporal information. To this end, we propose an adapter-based plug-and-play module for the backbone of 3D scene perception model, which constructs memory to cache and aggregate the extracted RGB-D features to empower offline models with temporal learning ability. Specifically, we propose a queued memory mechanism to cache the supporting point cloud and image features. Then we devise aggregation modules which directly perform on the memory and pass temporal information to current frame. We further propose 3D-to-2D adapter to enhance image features with strong global context. Our adapters can be easily inserted into mainstream offline architectures of different tasks and significantly boost their performance on online tasks. Extensive experiments on ScanNet and SceneNN datasets demonstrate our approach achieves leading performance on three 3D scene perception tasks compared with state-of-the-art online methods by simply finetuning existing offline models, without any model and task-specific designs. \href{https://xuxw98.github.io/Online3D/}{Project page}.

Shengjun Zhang, Xin Fei, Fangfu Liu et al.

3D Gaussian Splatting (3DGS) has demonstrated impressive novel view synthesis performance. While conventional methods require per-scene optimization, more recently several feed-forward methods have been proposed to generate pixel-aligned Gaussian representations with a learnable network, which are generalizable to different scenes. However, these methods simply combine pixel-aligned Gaussians from multiple views as scene representations, thereby leading to artifacts and extra memory cost without fully capturing the relations of Gaussians from different images. In this paper, we propose Gaussian Graph Network (GGN) to generate efficient and generalizable Gaussian representations. Specifically, we construct Gaussian Graphs to model the relations of Gaussian groups from different views. To support message passing at Gaussian level, we reformulate the basic graph operations over Gaussian representations, enabling each Gaussian to benefit from its connected Gaussian groups with Gaussian feature fusion. Furthermore, we design a Gaussian pooling layer to aggregate various Gaussian groups for efficient representations. We conduct experiments on the large-scale RealEstate10K and ACID datasets to demonstrate the efficiency and generalization of our method. Compared to the state-of-the-art methods, our model uses fewer Gaussians and achieves better image quality with higher rendering speed.

Haoyu Wu, Diankun Wu, Tianyu He et al.

Videos inherently represent 2D projections of a dynamic 3D world. However, our analysis suggests that video diffusion models trained solely on raw video data often fail to capture meaningful geometric-aware structure in their learned representations. To bridge this gap between video diffusion models and the underlying 3D nature of the physical world, we propose Geometry Forcing, a simple yet effective method that encourages video diffusion models to internalize latent 3D representations. Our key insight is to guide the model's intermediate representations toward geometry-aware structure by aligning them with features from a pretrained geometric foundation model. To this end, we introduce two complementary alignment objectives: Angular Alignment, which enforces directional consistency via cosine similarity, and Scale Alignment, which preserves scale-related information by regressing unnormalized geometric features from normalized diffusion representation. We evaluate Geometry Forcing on both camera view-conditioned and action-conditioned video generation tasks. Experimental results demonstrate that our method substantially improves visual quality and 3D consistency over the baseline methods. Project page: https://GeometryForcing.github.io.

Hanyang Wang, Fangfu Liu, Jiawei Chi et al.

Recovering 3D scenes from sparse views is a challenging task due to its inherent ill-posed problem. Conventional methods have developed specialized solutions (e.g., geometry regularization or feed-forward deterministic model) to mitigate the issue. However, they still suffer from performance degradation by minimal overlap across input views with insufficient visual information. Fortunately, recent video generative models show promise in addressing this challenge as they are capable of generating video clips with plausible 3D structures. Powered by large pretrained video diffusion models, some pioneering research start to explore the potential of video generative prior and create 3D scenes from sparse views. Despite impressive improvements, they are limited by slow inference time and the lack of 3D constraint, leading to inefficiencies and reconstruction artifacts that do not align with real-world geometry structure. In this paper, we propose VideoScene to distill the video diffusion model to generate 3D scenes in one step, aiming to build an efficient and effective tool to bridge the gap from video to 3D. Specifically, we design a 3D-aware leap flow distillation strategy to leap over time-consuming redundant information and train a dynamic denoising policy network to adaptively determine the optimal leap timestep during inference. Extensive experiments demonstrate that our VideoScene achieves faster and superior 3D scene generation results than previous video diffusion models, highlighting its potential as an efficient tool for future video to 3D applications. Project Page: https://hanyang-21.github.io/VideoScene

Shengjun Zhang, Jinzhao Li, Xin Fei et al.

In this paper, we propose Scene Splatter, a momentum-based paradigm for video diffusion to generate generic scenes from single image. Existing methods, which employ video generation models to synthesize novel views, suffer from limited video length and scene inconsistency, leading to artifacts and distortions during further reconstruction. To address this issue, we construct noisy samples from original features as momentum to enhance video details and maintain scene consistency. However, for latent features with the perception field that spans both known and unknown regions, such latent-level momentum restricts the generative ability of video diffusion in unknown regions. Therefore, we further introduce the aforementioned consistent video as a pixel-level momentum to a directly generated video without momentum for better recovery of unseen regions. Our cascaded momentum enables video diffusion models to generate both high-fidelity and consistent novel views. We further finetune the global Gaussian representations with enhanced frames and render new frames for momentum update in the next step. In this manner, we can iteratively recover a 3D scene, avoiding the limitation of video length. Extensive experiments demonstrate the generalization capability and superior performance of our method in high-fidelity and consistent scene generation.

Yang Chen, Yueqi Duan, Runzhong Zhang et al.

In this paper, we propose an adaptive margin contrastive learning method for 3D point cloud semantic segmentation, namely AMContrast3D. Most existing methods use equally penalized objectives, which ignore per-point ambiguities and less discriminated features stemming from transition regions. However, as highly ambiguous points may be indistinguishable even for humans, their manually annotated labels are less reliable, and hard constraints over these points would lead to sub-optimal models. To address this, we design adaptive objectives for individual points based on their ambiguity levels, aiming to ensure the correctness of low-ambiguity points while allowing mistakes for high-ambiguity points. Specifically, we first estimate ambiguities based on position embeddings. Then, we develop a margin generator to shift decision boundaries for contrastive feature embeddings, so margins are narrowed due to increasing ambiguities with even negative margins for extremely high-ambiguity points. Experimental results on large-scale datasets, S3DIS and ScanNet, demonstrate that our method outperforms state-of-the-art methods.

Boyuan Wang, Xiaofeng Wang, Yongkang Li et al.

Reconstructing non-rigid objects with physical plausibility remains a significant challenge. Existing approaches leverage differentiable rendering for per-scene optimization, recovering geometry and dynamics but requiring expensive tuning or manual annotation, which limits practicality and generalizability. To address this, we propose ReconPhys, the first feedforward framework that jointly learns physical attribute estimation and 3D Gaussian Splatting reconstruction from a single monocular video. Our method employs a dual-branch architecture trained via a self-supervised strategy, eliminating the need for ground-truth physics labels. Given a video sequence, ReconPhys simultaneously infers geometry, appearance, and physical attributes. Experiments on a large-scale synthetic dataset demonstrate superior performance: our method achieves 21.64 PSNR in future prediction compared to 13.27 by state-of-the-art optimization baselines, while reducing Chamfer Distance from 0.349 to 0.004. Crucially, ReconPhys enables fast inference (<1 second) versus hours required by existing methods, facilitating rapid generation of simulation-ready assets for robotics and graphics.

Shengjun Zhang, Xin Fei, Yueqi Duan

Point clouds captured by different sensors such as RGB-D cameras and LiDAR possess non-negligible domain gaps. Most existing methods design different network architectures and train separately on point clouds from various sensors. Typically, point-based methods achieve outstanding performances on even-distributed dense point clouds from RGB-D cameras, while voxel-based methods are more efficient for large-range sparse LiDAR point clouds. In this paper, we propose geometry-to-voxel auxiliary learning to enable voxel representations to access point-level geometric information, which supports better generalisation of the voxel-based backbone with additional interpretations of multi-sensor point clouds. Specifically, we construct hierarchical geometry pools generated by a voxel-guided dynamic point network, which efficiently provide auxiliary fine-grained geometric information adapted to different stages of voxel features. We conduct experiments on joint multi-sensor datasets to demonstrate the effectiveness of GeoAuxNet. Enjoying elaborate geometric information, our method outperforms other models collectively trained on multi-sensor datasets, and achieve competitive results with the-state-of-art experts on each single dataset.

Chensheng Dai, Shengjun Zhang, Min Chen et al.

3D Gaussian Splatting (3DGS) has demonstrated impressive performance in 3D scene reconstruction. Beyond novel view synthesis, it shows great potential for multi-view surface reconstruction. Existing methods employ optimization-based reconstruction pipelines that achieve precise and complete surface extractions. However, these approaches typically require dense input views and high time consumption for per-scene optimization. To address these limitations, we propose SurfelSplat, a feed-forward framework that generates efficient and generalizable pixel-aligned Gaussian surfel representations from sparse-view images. We observe that conventional feed-forward structures struggle to recover accurate geometric attributes of Gaussian surfels because the spatial frequency of pixel-aligned primitives exceeds Nyquist sampling rates. Therefore, we propose a cross-view feature aggregation module based on the Nyquist sampling theorem. Specifically, we first adapt the geometric forms of Gaussian surfels with spatial sampling rate-guided low-pass filters. We then project the filtered surfels across all input views to obtain cross-view feature correlations. By processing these correlations through a specially designed feature fusion network, we can finally regress Gaussian surfels with precise geometry. Extensive experiments on DTU reconstruction benchmarks demonstrate that our model achieves comparable results with state-of-the-art methods, and predict Gaussian surfels within 1 second, offering a 100x speedup without costly per-scene training.

Fangfu Liu, Hao Li, Jiawei Chi et al.

Recovering 3D structures with open-vocabulary scene understanding from 2D images is a fundamental but daunting task. Recent developments have achieved this by performing per-scene optimization with embedded language information. However, they heavily rely on the calibrated dense-view reconstruction paradigm, thereby suffering from severe rendering artifacts and implausible semantic synthesis when limited views are available. In this paper, we introduce a novel generative framework, coined LangScene-X, to unify and generate 3D consistent multi-modality information for reconstruction and understanding. Powered by the generative capability of creating more consistent novel observations, we can build generalizable 3D language-embedded scenes from only sparse views. Specifically, we first train a TriMap video diffusion model that can generate appearance (RGBs), geometry (normals), and semantics (segmentation maps) from sparse inputs through progressive knowledge integration. Furthermore, we propose a Language Quantized Compressor (LQC), trained on large-scale image datasets, to efficiently encode language embeddings, enabling cross-scene generalization without per-scene retraining. Finally, we reconstruct the language surface fields by aligning language information onto the surface of 3D scenes, enabling open-ended language queries. Extensive experiments on real-world data demonstrate the superiority of our LangScene-X over state-of-the-art methods in terms of quality and generalizability. Project Page: https://liuff19.github.io/LangScene-X.