3D instance segmentation is a pivotal yet demanding problem in scene understanding and perception. Existing approaches still fall short of fully exploiting geometric and semantic cues, and the weak coupling between these modalities often leads to over- and under-segmentation artifacts. In this paper, we present a novel framework that couples structural perception with semantic learning to exploit their complementary strengths for 3D instance segmentation. Our framework centers on instance mask prediction and augments it with semantic classification and bounding-box regression as auxiliary objectives. First, a knowledge embedding module initializes instance queries alongside point-level structural features, stabilizing training and accelerating convergence. Second, a two-stage refinement module iteratively updates the instance queries and their associated point features, strengthening the network's ability to align each instance with its constituent points. Finally, a joint mask module fuses geometric and semantic cues, capitalizing on their synergy to improve instance mask accuracy. Extensive evaluation on ScanNetV2, ScanNet200, and S3DIS benchmarks demonstrate that our method achieves state-of-the-art performance for 3D instance segmentation. Furthermore, our underlying architecture generalizes naturally to 3D object detection, where it achieves competitive performance. Source code is available at https://github.com/tianhuixiao12138/BGS.

Computing geodesic distances on 3D surfaces is fundamental to many tasks in 3D vision and geometry processing, with deep connections to tasks such as shape correspondence. Recent learning-based methods achieve strong performance but rely on large 3D backbones, leading to high memory usage and latency, which limit their use in interactive or resource-constrained settings. We introduce LiteGE, a lightweight approach that constructs compact, category-aware shape descriptors by applying PCA to unsigned distance field (UDFs) samples at informative voxels. This descriptor is efficient to compute and removes the need for high-capacity networks. LiteGE remains robust on sparse point clouds, supporting inputs with as few as 300 points, where prior methods fail. Extensive experiments show that LiteGE reduces memory usage and inference time by up to 300x compared to existing neural approaches. In addition, by exploiting the intrinsic relationship between geodesic distance and shape correspondence, LiteGE enables fast and accurate shape matching. Our method achieves up to 1000x speedup over state-of-the-art mesh-based approaches while maintaining comparable accuracy on non-isometric shape pairs, including evaluations on point-cloud inputs.

Point cloud data has been extensively studied due to its compact form and flexibility in representing complex 3D structures. The ability of point cloud data to accurately capture and represent intricate 3D geometry makes it an ideal choice for a wide range of applications, including computer vision, robotics, and autonomous driving, all of which require an understanding of the underlying spatial structures. Given the challenges associated with annotating large-scale point clouds, self-supervised point cloud representation learning has attracted increasing attention in recent years. This approach aims to learn generic and useful point cloud representations from unlabeled data, circumventing the need for extensive manual annotations. In this paper, we present a comprehensive survey of self-supervised point cloud representation learning using DNNs. We begin by presenting the motivation and general trends in recent research. We then briefly introduce the commonly used datasets and evaluation metrics. Following that, we delve into an extensive exploration of self-supervised point cloud representation learning methods based on these techniques. Finally, we share our thoughts on some of the challenges and potential issues that future research in self-supervised learning for pre-training 3D point clouds may encounter.

With the growing demand for real-world 3-D understanding, learning effective representations of 3-D data has become increasingly important for tasks such as shape classification, model retrieval, scene reconstruction, and point cloud completion. Although previous work has explored self-supervised learning within individual modalities (e.g., point clouds or images), the potential of multi-modal supervision remains largely underexplored due to the lack of aligned and scalable training signals. In this work, we present DR-Point, a tri-modal pre-training framework that jointly learns from RGB images, depth maps, and 3-D point clouds to build a unified embedding space across modalities. By enforcing cross-modal consistency among RGB-depth-point triplets, DR-Point achieves effective 2-D-3-D feature alignment without manual annotations. A differentiable rendering module further enhances geometric fidelity by synthesizing depth cues and refining structural details in reconstructed point clouds. Extensive experiments on benchmarks demonstrate that DR-Point consistently outperforms state-of-the-art self-supervised methods on 3-D classification, segmentation, and completion. These results highlight the advantages of multi-modal pre-training for unified 3-D understanding and its potential to benefit a wide range of vision and graphics applications.

Existing image restoration methods primarily rely on the posterior distribution of natural images but are often limited by their dependence on known degradations and supervised training. To this end, we propose Generative Diffusion Prior (GDP), an unsupervised sampling-based framework that effectively models posterior distributions for image and video restoration. GDP utilizes a single pre-trained denoising diffusion probabilistic model (DDPM) to solve a wide range of linear, non-linear, and blind inverse problems without explicit degradation assumptions. Specifically, GDP systematically explores a conditional guidance protocol, which proves more practical and effective than conventional methods of adding guidance. Furthermore, GDP incorporates a degradation model optimization mechanism during the denoising process, enabling blind image restoration. Besides, we introduce a patch-based strategy, allowing GDP to handle images of arbitrary resolution. We extensively evaluate GDP on multiple image and video restoration tasks, including super-resolution, deblurring, inpainting, and colorization, as well as more challenging applications such as low-light enhancement, HDR recovery, and LDR video enhancement. Experimental results demonstrate that GDP outperforms leading unsupervised methods across diverse benchmarks in both reconstruction accuracy and perceptual quality, while demonstrating robust generalization to images and videos of any size. Our project page at https://generativediffusionprior.github.io/.

Computer-aided design (CAD) plays a central role in engineering, manufacturing, and 3D content creation, but producing precise CAD models directly from natural language is still difficult. We propose Text2CSG, a two-stage framework for producing high-quality CAD models from text. Our approach uses Constructive Solid Geometry (CSG) as an intermediate representation to capture the structured and hierarchical nature of CAD construction. Unlike prior approaches based on sketch-extrusion sequences, Text2CSG exploits the Boolean compositional logic of CSG trees to represent complex shapes. The system consists of two primary components: (1) a geometry decoder that reconstructs 3D shapes from CSG programs, and (2) a structure-aware Transformer conditioned on high-level semantics extracted from a frozen large language model (LLM). By decoupling semantic understanding from geometric generation, our approach benefits from pre-trained LLMs without requiring fine-tuning. Experiments on a newly constructed CSG dataset demonstrate that Text2CSG achieves higher geometric accuracy and stronger structural consistency than existing methods. Furthermore, the utilization of structured CSG programs makes the generation process interpretable and editable, enabling flexible text-driven CAD design. We will release the source code and dataset.

The Medial Axis Transform (MAT) is a complete shape descriptor capable of reconstructing the geometry of the original domain. A high-quality MAT should not only facilitate high-fidelity reconstruction but also capture structural features – for instance, by aligning the MAT boundary with the locus of rolling ball centers within fillet regions. However, computing such an ideal MAT remains a significant challenge, particularly when the input is a discrete triangle mesh. In this paper, we follow the established technical pipeline of initializing the MAT via a 3D Voronoi diagram of surface samples and subsequently simplifying the Voronoi structure through a QEM-like scheme. Our key insight is to explicitly track the correspondence between MAT vertices and surface regions throughout the progressive simplification process, ensuring that the resulting MAT triangles accurately reflect the intrinsic symmetries between surface patches. We translate these geometric requirements into a suite of priority control strategies that govern the sequencing of edge collapses. Through extensive evaluation against state-of-the-art MAT algorithms, we validate the strong performance of our approach regarding runtime efficiency, structural alignment, boundary regularity, triangle quality, and robustness to noise. Our resulting MATs remain highly expressive for both articulated shapes and CAD models, even under extreme simplification – effectively capturing the global structure of complex geometries with only a few hundred vertices.

Giving machines the ability to infer the complete 3D geometry and semantics of complex scenes is crucial for many downstream tasks, such as decision-making and planning. Vision-centric Semantic Scene Completion (SSC) has emerged as a trendy 3D perception paradigm due to its compatibility with task properties, low cost, and rich visual cues. Despite impressive results, current approaches inevitably suffer from problems such as depth errors or depth ambiguities during the 2D-to-3D transformation process. To overcome these limitations, in this paper, we first introduce an Optical Flow-Guided (OFG) Depth-Net that leverages the strengths of pretrained depth estimation models, while incorporating optical flow images to improve depth prediction accuracy in regions with significant depth changes. Then, we propose a depth ambiguity-mitigated feature lifting strategy that implements deformable cross-attention in 3D pixel space to avoid depth ambiguities caused by the projection process from 3D to 2D and further enhances the effectiveness of feature updating through the utilization of prior mask indices. Moreover, we customize two subnetworks: a residual voxel network and a sparse UNet, to enhance the network’s geometric prediction capabilities and ensure consistent semantic reasoning across varying scales. By doing so, our method achieves performance improvements over state-of-the-art methods on the SemanticKITTI, SSCBench-KITTI-360 and Occ3D-nuScene benchmarks.

Surface parameterization is a fundamental geometry processing task, laying the foundations for the visual presentation of 3D assets and numerous downstream shape analysis scenarios. Conventional parameterization approaches demand high-quality mesh triangulation and are restricted to certain simple topologies unless additional surface cutting and decomposition are provided. In practice, the optimal configurations (e.g., type of parameterization domains, distribution of cutting seams, number of mapping charts) may vary drastically with different surface structures and task characteristics, thus requiring more flexible and controllable processing pipelines. To this end, this paper introduces FlexPara, an unsupervised neural optimization framework to achieve both global and multi-chart surface parameterizations by establishing point-wise mappings between 3D surface points and adaptively-deformed 2D UV coordinates. We ingeniously design and combine a series of geometrically-interpretable sub-networks, with specific functionalities of cutting, deforming, unwrapping, and wrapping, to construct a bi-directional cycle mapping framework for global parameterization without the need for manually specified cutting seams. Furthermore, we construct a multi-chart parameterization framework with adaptively-learned chart assignment. Extensive experiments demonstrate the universality, superiority, and inspiring potential of our neural surface parameterization paradigm.

High-fidelity surface reconstruction from multi-view images is a core problem in 3D computer vision. While neural implicit surfaces like SDFs offer smooth geometry, they are often bottlenecked by the computational intensity of volume rendering. Conversely, 3D Gaussian Splatting (3DGS) provides rapid training but lacks geometry continuity, often leading to fragmented surfaces. This paper presents a novel framework that integrates Signed Distance Fields directly into the splatting pipeline. By leveraging the continuous nature of SDFs to regularize Gaussian primitives, our method effectively fills geometric holes and suppresses noise inherent in sparse point clouds. Unlike hybrid approaches that rely on heavy volumetric sampling, our approach utilizes the efficiency of splatting to achieve faster convergence. Extensive evaluations demonstrate that our method produces high-quality surfaces with significantly fewer primitives, offering a more compact and efficient representation for both indoor and outdoor environments. The source code is available at https://github.com/xubaixinxbx/Gsurf.

Neural implicit methods reconstruct surfaces from point clouds by learning a continuous function, usually signed or unsigned distances. For unsigned distance field (UDF) reconstruction, many self-supervised methods typically estimate distance through local, gradient-guided displacement regression from query points to the surface. Because these corrections are local, they can become unstable near the zero level set and accumulate errors under sparse or non-uniform sampling. We address this limitation by reformulating UDF learning as a geometrically constrained trajectory-evolution problem, shifting from static local regression to global path-consistent modeling. Specifically, we construct a bidirectional linear flow that promotes shortest-path trajectories between surface samples and query points. Under this formulation, distance prediction is interpreted as continuous state evolution along deterministic trajectories, inherently preserving geometric consistency and mitigating local error accumulation. This perspective connects UDF estimation to neural ODEs, where geometry is recovered through continuous dynamics, with the learned transport following structured trajectories induced by an implicit geometric field. To address sparse and non-uniform sampling, we further introduce a flow-guided iterative densification strategy that progressively upgrades shortest-path modeling from discrete point-to-point approximations to accurate point-to-surface projections, enabling implicit geometric completion. Extensive experiments on synthetic benchmarks and real-world scans demonstrate that our method achieves state-of-the-art performance in both reconstruction accuracy and robustness.

Diffusion-based generative models have achieved remarkable success in image restoration by learning to iteratively refine noisy data toward clean signals. Inspired by this progress, recent efforts have begun exploring their potential in 3D domains. However, applying diffusion models to point cloud denoising introduces several challenges. Unlike images, clean and noisy point clouds are characterized by structured displacements. As a result, it is unsuitable to establish a transform mapping in the forward phase by diffusing Gaussian noise, as this approach disregards the inherent geometric relationship between the point sets. Furthermore, the stochastic nature of Gaussian noise introduces additional complexity, complicating geometric reasoning and hindering surface recovery during the reverse denoising process. In this paper, we introduce a deterministic noise-free diffusion framework that formulates point cloud denoising as a two-phase residual diffusion process. In the forward phase, directional residuals are injected into clean surfaces to construct a degradation trajectory that encodes both local displacements and their global evolution. In the reverse phase, a U-Net-based network iteratively estimates and removes these residuals, effectively retracing the degradation path backward to recover the underlying surface. By decomposing the denoising task into directional residual computation and sequential refinement, our method enables faithful surface recovery while mitigating common artifacts such as over-smoothing and under-smoothing. Extensive experiments on synthetic and real-world datasets demonstrate that our method achieves state-of-the-art performance in both quantitative metrics and visual quality.

3D object affordance grounding aims to identify regions on 3D objects that support human-object interaction (HOI), a capability essential to embodied visual reasoning. However, most existing approaches rely on static visual or textual cues, neglecting that affordances are inherently defined by dynamic actions. As a result, they often struggle to localize the true contact regions involved in real interactions. We take a different perspective. Humans learn how to use objects by observing and imitating actions, not just by examining shapes. Motivated by this intuition, we introduce video-guided 3D affordance grounding, which leverages dynamic interaction sequences to provide functional supervision. To achieve this, we propose VAGNet, a framework that aligns video-derived interaction cues with 3D structure to resolve ambiguities that static cues cannot address. To support this new setting, we introduce PVAD, the first HOI video-3D pairing affordance dataset, providing functional supervision unavailable in prior works. Extensive experiments on PVAD show that VAGNet achieves state-of-the-art performance, significantly outperforming static-based baselines. The code and dataset will be open publicly.

Recent advances in multi-view 3D reconstruction, especially neural implicit surface methods, can recover high-quality geometry by representing shape with signed distance functions (SDFs). However, reconstructing complex objects with severe self-occlusion remains difficult when all images are captured under a single object pose, because the observed views often provide incomplete spatial coverage. We propose PoseFusion, a multi-pose reconstruction framework that fuses neural implicit surfaces independently learned from different object poses into a unified 3D representation. PoseFusion follows a two-stage registration-and-fusion pipeline. First, we extract an oriented bounding box (OBB) from the mesh derived from each pose-specific SDF and use the OBBs, SDF samples, and multi-view image features to estimate a coarse inter-pose alignment. Second, we refine the alignment using image-and SDF-guided correspondences: cross-pose image matches are lifted to 3D using the learned SDFs and used to iteratively optimize the relative transformations. To facilitate evaluation, we construct a dataset of synthetic and real-world objects with complex geometry and strong self-occlusion, captured across multiple poses with calibrated multi-view images. Experiments show that PoseFusion is robust under challenging capture conditions and consistently produces high-fidelity reconstructions from multi-pose, multi-view inputs. Code and dataset are publicly available at https://raining00.github.io/PoseFusion-page/.

Recent advances in 3D Gaussian Splatting (3DGS) have significantly advanced indoor scene reconstruction, a key capability for AR/VR, robotics, and autonomous systems. However, existing methods that rely on 2D monocular priors often produce artifacts such as holes and unnatural protrusions due to insufficient or weak constraints. To address this limitation, we propose a geometry-semantics co-regularization framework that jointly optimizes geometry and semantics within 3DGS. On the geometric side, we introduce a multi-level geometric optimization strategy with two key contributions. First, inspired by neural implicit surface representations, we design a dual implicit architecture that integrates Implicit Moving Least Squares (IMLS) and Signed Distance Functions (SDFs). This structure enables mutual reinforcement between the two representations via local geometric approximation and global normal constraints, improving fine-grained surface modeling. Second, we develop a view-dependent normal consistency constraint that combines 2D observations with SDF gradients to refine surface accuracy. On the semantic side, we develop a multi-view semantic consistency supervision to regularize the semantic distributions of Gaussian primitives, ensuring cross-view consistency for Gaussians corresponding to the same semantic category or instance. Extensive experiments on the Replica and MuSHRoom datasets demonstrate that our approach achieves state-of-the-art geometric reconstruction quality and rendering fidelity. Check out the project page https://sgo-gauss.github.io for the source code and more results.

Reconstructing realistic 3D human avatars from monocular videos is a challenging task due to the limited geometric information and complex non-rigid motion involved. We present MonoCloth, a new method for reconstructing and animating clothed human avatars from monocular videos. To overcome the limitations of monocular input, we introduce a part-based decomposition strategy that separates the avatar into body, face, hands, and clothing. This design reflects the varying levels of reconstruction difficulty and deformation complexity across these components. Specifically, we focus on detailed geometry recovery for the face and hands. For clothing, we propose a dedicated cloth simulation module that captures garment deformation using temporal motion cues and geometric constraints. Experimental results demonstrate that MonoCloth improves both visual reconstruction quality and animation realism compared to existing methods. Furthermore, thanks to its part-based design, MonoCloth also supports additional tasks such as clothing transfer, underscoring its versatility and practical utility.

3D surface cutting and UV unwrapping are fundamental problems in computer graphics. Traditional geometric optimization methods mainly focus on reducing parameterization distortion, but they often overlook visual semantic coherence in seam layouts. Recent autoregressive generative methods improve semantic coherence, yet limited perception of mesh topology often causes inaccurate local cuts. To address these limitations, we introduce SeamFlow, a novel generative framework for 3D surface cutting. We reformulate the discrete mesh-cutting problem as continuous flow matching in a high-dimensional edge-probability space. Through continuous relaxation, SeamFlow learns a deterministic mapping from a Gaussian prior to a target seam-probability distribution. An evolution network couples local topological tokens with global shape priors and guides smooth probability flow through Ordinary Differential Equation solving. Compared with existing autoregressive generative frameworks, SeamFlow improves topology awareness through edge tokenization while eliminating both 3D spatial projection errors and artificial sequential-order bias. Extensive experiments demonstrate that SeamFlow achieves exceptional semantic coherence and remarkably low parameterization distortion. The project page is https://meshy-dev.github.io/seamflow.

Point cloud data are ubiquitous in 3D vision applications, yet they often suffer from noise, sparsity, and incompleteness, which significantly limit the performance of downstream tasks. To address these challenges, we propose LatentFlow3D, a unified framework for point cloud denoising, upsampling, and completion through latent space modeling with invertible neural networks. For denoising, we disentangle noise from clean geometry by projecting them onto separate channels in latent space, enabling the recovery of intrinsic clean latent codes. For upsampling, we formulate the task as interpolation in latent space, where adaptive weights are learned from local geometric context. For completion, we exploit geometric symmetry by applying affine transformations in latent space to infer the latent codes of missing regions, which are then fused with partial inputs to reconstruct complete shapes. To establish a bijective and expressive mapping between point clouds and latent representations, we employ invertible residual networks, augmented with a feature injection module that integrates rich local geometric features. Extensive experiments on widely used benchmarks demonstrate that LatentFlow3D achieves SOTA performance across all three tasks.

Neural Signed Distance Functions (SDFs) excel at reconstructing watertight manifolds but fail on thin structures and open boundaries due to strict inside--outside constraints. Conversely, Unsigned Distance Fields (UDFs) accommodate general geometries but suffer from gradient singularities at the zero-level set, hindering optimization and extraction. We introduce Metric--Phase Fields (MPFs), a decoupled implicit representation that separates metric proximity from topological phase. Given an unoriented point cloud, MPFs learn (i) an unsigned metric field $r$ and (ii) a smooth phase field $θ$, for which we derive a bounded phase indicator $P=\tanh(βθ)$ that provides soft inside--outside cues where they are meaningful. We couple the two fields via a gated-metric formulation with a residual phase injection to obtain a signed implicit function with stable near-surface gradients. The phase coefficient $β$ is learnable, allowing MPFs to adaptively control the sharpness of the phase transition and the degree of saturation of the soft sign indicator. Experiments on both synthetic and scanned thin-shell and thin-plate shapes demonstrate that MPFs preserve thin and layered structures more faithfully than recent SDF-based methods, while also enabling more robust training and more reliable surface extraction than UDF-based approaches. Check out \href{https://github.com/JIAYI-Scarlett/ICML2026-MPF}{MPFs-GitHub} for source code and test models.

Human comprehension of multimodal information follows a hierarchical cognitive process that evolves from perception to integrated understanding. Inspired by this mechanism, we reinterpret multi-modal relation extraction (MRE) as a cognitive progression rather than a flat feature fusion task as in the existing methods. We propose a Cognitive Hierarchy Framework (CHF) that models multimodal reasoning through sequential stages: perception, association, reasoning, self-regulation and comprehension. Each stage incrementally aligns and refines cross-modal representations, leading to interpretable and efficient relational understanding without incorporating external structured knowledge or explicit relational supervision. Experiments on two datasets show that CHF achieves state-of-the-art performance, improving F1 by 2.37% on MNRE and 1.25% on MORE, while reducing fusion parameters by 62%. Ablation and visualization analyses further demonstrate how cross-modal consistency is gradually established across stages. These results suggest that incorporating hierarchical cognitive principles provides a promising direction for human-like multimodal reasoning.