CVPR 2026 论文列表

360deg panoramic images are increasingly used in VR, autonomous driving, and robotics for holistic scene understanding. However, current Vision-Language Models (VLMs) struggle with 3D spatial reasoning on Equirectangular Projection (ERP) images due to geometric distortion and limited 3D supervision. We introduce PanoEnv, a large-scale VQA benchmark built from synthetic 3D environments, containing 14.8K questions across five categories (e.g., relative position, attribute comparison) grounded in accurate 3D annotations--depth, segmentation, and bounding boxes. Benchmarking 14 state-of-the-art VLMs reveals limited 3D understanding, achieving only 49.34% overall and 8.36% on open-ended (OE) questions. To enhance 3D reasoning, we propose a reinforcement learning post-training framework based on Group Relative Policy Optimization (GRPO) with a ground-truth-guided reward combining five geometry-aware strategies (e.g., distance tolerance, spatial consistency). A two-stage curriculum further mitigates catastrophic forgetting: Stage 1 trains on structured tasks (T/F, MCQ), and Stage 2 fine-tunes on mixed OE data for generalization. Our 7B model sets a new SoTA performance, improving total accuracy to 52.93% (+3.59%) and OE accuracy to 14.83% while maintaining structured-task performance. It also achieves top semantic scores (Q-Score 6.24, P-Score 5.95), surpassing 32B models. These results demonstrate that PanoEnv-QA and our curriculum-based RL framework effectively instill 3D spatial intelligence in VLMs for omnidirectional perception.

Multimodal large reasoning models (MLRMs) often suffer from hallucinations that stem not only from insufficient visual grounding but also from imbalanced allocation between perception and reasoning processes. Building upon recent interpretability findings suggesting a staged division of attention across layers, we analyze how this functional misalignment leads to two complementary failure modes: perceptual bias in shallow layers and reasoning drift in deeper layers. To alleviate these issues, we propose Functional Head Identification and Class-Conditioned Rescaling, a lightweight, training-free plugin that identifies perception- and reasoning-oriented heads and adaptively rebalances their layerwise contributions. Our method improves reasoning consistency and visual faithfulness without retraining or any architectural modification. Evaluations across three representative MLRMs and five multimodal reasoning benchmarks show an average 4.2 percentage-point gain, with less than 1% additional computation and only 9% baseline latency. Beyond empirical improvements, our study provides an interpretable perspective on regulating cross-layer functional dynamics to enhance the reliability of multimodal reasoning.

Color constancy aims to keep object colors consistent under varying illumination. Cross-camera generalization in color constancy remains challenging because learning-based models often overfit to the color response characteristics of the training camera, resulting in degraded performance on images captured by other cameras. We propose VLM-CC, a feedback-guided framework that formulates color constancy as an iterative refinement process. Instead of directly estimating the illuminant from raw input, VLM-CC performs iterative correction driven by vision-language model (VLM)-based evaluation. At each iteration, the image is white-balanced using the current estimate and converted to pseudo-sRGB. A lightweight LoRA-tuned VLM then assesses the corrected image, identifying the dominant residual color cast and providing qualitative feedback. This feedback is mapped to a residual illumination direction (red, green, or blue) and used to update the illuminant estimate until convergence. Our key idea is to reframe color constancy as an iterative perceptual feedback problem, leveraging VLM evaluation instead of direct RGB regression. By replacing direct RGB estimation with VLM-guided perceptual feedback, VLM-CC achieves state-of-the-art robustness in cross-camera color constancy across multiple datasets.

Prototype or region-attention modules have recently improved medical image segmentation but still suffer from two fundamental limitations: 1) they represent each semantic concept as a point or isotropic region, failing to capture the inherently anisotropic geometry of real feature distributions; and 2) many rely on non-differentiable clustering or one-way kernel weighting, which restricts their ability to form coherent region-level representations. We address these issues with the Anisotropic Differentiable Granular-Ball (AD-GBC) module, which generalizes prototypes into learnable geometric regions parameterized by a center and an anisotropic vector scale. AD-GBC aggregates local features into region-level semantics and redistributes the refined representation back to pixels in a fully differentiable manner, enabling geometry-aware refinement within modern UNet-style architectures. Two geometric regularizers, a Wasserstein-based diversity loss and a scale consistency loss, mitigate center collapse and encourage stable, well-formed region geometry.AD-GBC yields consistent improvements across four widely used medical segmentation benchmarks (BUSI, GlaS, CVC-ClinicDB, ISIC17) when integrated into two strong backbones (Rolling-UNet and U-KAN), demonstrating that the proposed geometric region formulation generalizes well across different imaging conditions. The code is available at https://github.com/SiaShen-dot/AD-GBC.

Distinct from attention-based compression methods, this paper presents an information uniqueness driven video compression framework, termed UniComp, which aims to maximize the information fidelity of video representations under constrained computational budgets. Starting from the information-theoretic perspective, we formulate the vision compression as an optimization problem that minimizes conditional entropy (reconstruction error) between retained and full tokens. To achieve this, we introduce the notion of information uniqueness to measure intrinsic redundancy among tokens to link with reconstruction error. Based on uniqueness, we design three modules--Frame Group Fusion, Token Allocation, and Spatial Dynamic Compression--that progressively perform semantic frame grouping, adaptive resource allocation, and fine-grained spatial compression. Extensive experiments demonstrate that UniComp consistently outperforms existing compression methods in preserving essential visual tokens under limited computational budgets, highlighting the pivotal role of information uniqueness in token compression efficacy.

Feed-forward 3D Gaussian Splatting (3DGS) enables efficient one-pass scene reconstruction, providing 3D representations for novel view synthesis without per-scene optimization. However, existing methods typically predict pixel-aligned primitives per-view, producing an excessive number of primitives in dense-view settings and offering no explicit control over the number of predicted Gaussians. To address this, we propose EcoSplat, the first efficiency-controllable feed-forward 3DGS framework that adaptively predicts the 3D representation for any given target primitive count at inference time. EcoSplat adopts a two-stage optimization process. The first stage is Pixel-aligned Gaussian Training (PGT) where our model learns initial primitive prediction. The second stage is Importance-aware Gaussian Finetuning (IGF) stage where our model learns rank primitives and adaptively adjust their parameters based on the target primitive count. Extensive experiments across multiple dense-view settings show that EcoSplat is robust and outperforms state-of-the-art methods under strict primitive-count constraints, making it well-suited for flexible downstream rendering tasks. Code and project page will be released.

Few-shot Class-Incremental Learning (FSCIL) requires learning new classes from very limited data while preventing catastrophic forgetting. Existing methods rely mainly on visual features and are prone to overfitting, while recent vision-language models (VLMs) offer better transferability but suppress fine-grained information due to contrastive feature decorrelation. Moreover, current FSCIL approaches often use static or fully optimizable prompts, making them either rigid or susceptible to semantic drift in incremental sessions. We introduce QR-Prompt, a residual-driven framework that leverages the visual-textual feature residual of VLMs to recover discriminative fine-grained cues missing from the contrastive space. To ensure stability, we propose Discriminative Subspace Quantization (DSQ), which builds a discrete memory of residual subspaces. To enable plasticity, a Hierarchical Prompt Encoder (HPE) and Prompt Composer (PC) transform these discrete codes into continuous, class-adaptive prompts for novel classes. We derive bounds relating DSQ codebook size to generalization and classification margin, and achieve consistent improvements over state-of-the-art FSCIL methods on CUB200, CIFAR100, and miniImageNet. Our results show that residual-based quantization combined with hierarchical prompt composition yields stable and expressive VLM adaptation for FSCIL.

We introduce the "single-life" learning paradigm, where we train a distinct vision model exclusively on egocentric videos captured by one individual. We leverage the multiple viewpoints naturally captured within a single life to learn a visual encoder in a self-supervised manner. Our experiments demonstrate three key findings. First, models trained independently on different lives develop a highly aligned geometric understanding. We demonstrate this by training visual encoders on distinct datasets each capturing a different life, both indoors and outdoors, as well as introducing a novel cross-attention-based metric to quantify the functional alignment of the internal representations developed by different models. Second, we show that single-life models learn generalizable geometric representations that effectively transfer to downstream tasks, such as depth estimation, in unseen environments. Third, we demonstrate that training on up to 30 hours from one week of the same person's life leads to comparable performance to training on 30 hours of diverse web data, highlighting the strength of single-life representation learning. Overall, our results establish that the shared structure of the world, both leads to consistency in models trained on individual lives, and provides a powerful signal for visual representation learning.

Computer-Aided Design (CAD) is essential in industrial design, but the complexity of traditional CAD modeling and workflows presents significant challenges for automating the generation of high-precision, editable CAD models. Existing methods, such as 3D reconstruction from sketches, often produce non-editable, approximate models that fall short of meeting the stringent requirements for precision and editability in industrial design. Moreover, the reliance on text or image-based inputs often requires significant manual annotation, limiting their scalability and applicability in industrial settings. To overcome these challenges, we propose the Heterogeneous Collaborative Multi-Expert Reinforcement Learning (CME-CAD) paradigm, a novel training paradigm for CAD code generation. Our approach integrates the complementary strengths of these models, facilitating collaborative learning and improving the model's ability to generate accurate, constraint-compatible, and fully editable CAD models. We introduce a two-stage training process: Multi-Expert Fine-Tuning (MEFT), and Multi-Expert Reinforcement Learning (MERL). Additionally, we present CADExpert, an open-source benchmark consisting of 17,299 instances, including orthographic projections with precise dimension annotations, expert-generated Chain-of-Thought (CoT) processes, executable CADQuery code, and rendered 3D models. The CADExpert dataset is publicly available at https://modelscope.cn/datasets/zhuofanChen/CADExpert.

A world model is an internal model that simulates how the world evolves. Given past observations and actions, it predicts the future physical state of both the embodied agent and its environment. Accurate world models are essential for enabling agents to think, plan, and reason effectively in complex, dynamic settings. However, existing world models often focus on random generation of open worlds, but neglect the need for high-fidelity modeling of deterministic scenarios (such as fixed-map mazes and static space robot navigation). In this work, we take a step toward building a truly accurate world model by addressing a fundamental yet open problem: constructing a model that can fully clone a deterministic world. 1) Through diagnostic experiment, we quantitatively demonstrate that high-fidelity cloning is feasible and the primary bottleneck for long-horizon fidelity is the geometric structure of the latent representation, not the dynamics model itself. 2) Building on this insight, we show that applying temporal contrastive learning principle as a geometric regularization can effectively curate a latent space that better reflects the underlying physical state manifold, demonstrating that contrastive constraints can serve as a powerful inductive bias for stable world modeling; we call this approach Geometrically-Regularized World Models (GRWM). At its core is a lightweight geometric regularization module that can be seamlessly integrated into standard autoencoders, reshaping their latent space to provide a stable foundation for effective dynamics modeling. By focusing on representation quality, GRWM offers a simple yet powerful pipeline for improving world model fidelity. Code is available at https://github.com/XiaFire/Clone_Deterministic_Environment.

Recent advances in 3D scene editing using NeRF and 3DGS enable high-quality static scene editing. In contrast, dynamic scene editing remains challenging, as methods that directly extend 2D diffusion models to 4D often produce motion artifacts, temporal flickering, and inconsistent style propagation. We introduce Catalyst4D, a framework that transfers high-quality 3D edits to dynamic 4D Gaussian scenes while maintaining spatial and temporal coherence. At its core, Anchor-based Motion Guidance (AMG) builds a set of structurally stable and spatially representative anchors from both original and edited Gaussians. These anchors serve as robust region-level references, and their correspondences are established via optimal transport to enable consistent deformation propagation without cross-region interference or motion drift. Complementarily, Color Uncertainty-guided Appearance Refinement (CUAR) preserves temporal appearance consistency by estimating per-Gaussian color uncertainty and selectively refining regions prone to occlusion-induced artifacts. Extensive experiments demonstrate that Catalyst4D achieves temporally stable, high-fidelity dynamic scene editing and outperforms existing methods in both visual quality and motion coherence.

Image generative models have become indispensable tools to yield exquisite high-resolution (HR) images for everyone, ranging from general users to professional designers. However, a desired outcome often requires generating a large number of HR images with different prompts and seeds, resulting in high computational cost for both users and service providers. Generating low-resolution (LR) images first could alleviate computational burden, but it is not straightforward how to generate LR images that are perceptually consistent with their HR counterparts. Here, we consider the task of generating high-fidelity LR images, called Previews, that preserve perceptual similarity of their HR counterparts for an efficient workflow, allowing users to identify promising candidates before generating the final HR image. We propose the commutator-zero condition to ensure the LR-HR perceptual consistency for flow matching models, leading to the proposed training-free solution with downsampling matrix selection and commutator-zero guidance. Extensive experiments show that our method can generate LR images with up to 33% computation reduction while maintaining HR perceptual consistency. When combined with existing acceleration techniques, our method achieves up to 3xspeedup. Moreover, our formulation can be extended to image manipulations, such as warping and translation, demonstrating its generalizability.

Vision-Language-Action (VLA) models are emerging as an important technology in autonomous driving, recognized for their sophisticated reasoning and interpretability. However, traditional VLA models often rely on image-to-text with Chain-of-Thought (CoT) reasoning, which converts sequential visual scenes into textual symbols, thereby under-utilizing spatial context in visual information. Existing autonomous driving systems using VLA models predict only a single sequence of waypoints as a trajectory considering a given command and multiple aspects. However, we suggest evaluating each sequence of waypoints to reveal the importance of the corresponding aspect. We introduce HybridDriveVLA, a VLA model that integrates visual Chain-of-Thought (V-Cot) reasoning and a proposed Tree-of-Thought (ToT)-inspired waypoint evaluation (ToT-evaluation). V-Cot reasoning anticipates future scenes, which serve as goals for ToT-evaluation. The ToT-evaluation generates and scores waypoints based each on safety, progress, and comfort aspects. The highest cumulated score of the waypoints based on the three aspects is optimal. To the best of our knowledge, we are the first to propose a unified method integrating both a CoT and a ToT approach in a VLA model. Experimental results demonstrated that HybridDriveVLA achieved strong performance on comfort, progress, and safety metrics with and average collision rate of 0.17% on the nuScenes benchmark, outperforming traditional VLA models.

Fully decentralized deep learning removes global servers and ensures local data privacy. However, Euclidean consensus, averaging weights, gradients or momentum, may degrade under the situations with non-i.i.d. (non-independent and identically distributed) data and client size imbalance. We propose a geometry-aware approach based on natural gradient variational inference. Clients communicate in the expectation parameter space of an exponential family, where simple linear mixing yields a forward Kullback--Leibler (KL) barycenter consensus. The aggregate is the model closest to all client distributions, aligning updates across heterogeneous sites and mitigating distribution shift. We further provide a lightweight decentralized Adam implementation where each client maintains a diagonal Gaussian posterior and operates entirely in the expectation space for both optimization and consensus. We prove convergence for convex losses on connected graphs. On CIFAR-100 and a medical image segmentation benchmark, our method substantially outperforms Euclidean-space consensus baselines under severe non-i.i.d. and client-imbalance cases, achieving around 20% accuracy gain on CIFAR-100, while matching the communication budget and improving training stability. The code is available at https://github.com/x-lu/Beyond-euclidean-gossip.

Aligning video generative models with human preferences remains challenging: current approaches rely on Vision-Language Models (VLMs) for reward modeling, but these models struggle to capture subtle temporal dynamics. We propose a fundamentally different approach: repurposing video generative models, which are inherently designed to model temporal structure, as reward models. We present the Generative-Transformer-based Self-Supervised Video Judge (\modelname), a novel evaluation model that transforms state-of-the-art video generation models into powerful temporally-aware reward models. Our key insight is that generative models can be reformulated as energy-based models (EBMs) that assign low energy to high-quality videos and high energy to degraded ones, enabling them to discriminate video quality with remarkable precision when trained via contrastive objectives. To prevent the model from exploiting superficial differences between real and generated videos, we design challenging synthetic negative videos through controlled latent-space perturbations: temporal slicing, feature swapping, and frame shuffling, which simulate realistic but subtle visual degradations. This forces the model to learn meaningful spatiotemporal features rather than trivial artifacts. \modelname achieves state-of-the-art performance on GenAI-Bench and MonteBench using only 30K human-annotations: 6xto 65xfewer than existing VLM-based approaches.

Deep learning models continue to scale, with some requiring more storage than many large-scale datasets. Thus, we introduce a new paradigm: Continual Distillation (CD), where a student learns sequentially from a stream of teacher models without retaining access to earlier teachers. CD faces two challenges: teacher training data is unavailable, and teachers have varying expertise. We show that external unlabeled data enables Unseen Knowledge Transfer (UKT), allowing the student to acquire information from domains not present in the training data, while known to the teacher. We also show that sequential distillation causes Unseen Knowledge Forgetting (UKF) when transferred knowledge is lost after training on later teachers. To better trade off between UKT and UKF, we propose Self External Data Distillation (SE2D), a method that preserves logits on external data to stabilize learning across heterogeneous teachers. Experiments on multiple benchmarks show that SE2D reduces UKF and improves cross-domain generalization. The code and implementation for this work are publicly available at: https://github.com/Nicolas1203/continual_distillation

Spike camera is a neuromorphic vision sensor with ultra-high temporal resolution, capable of capturing fast-moving scenes by firing a stream of binary spikes. However, its relatively low spatial resolution limits the acquisition of fine-grained visual details, motivating research on spike camera super resolution (SCSR). Existing SCSR methods typically operate on fixed-length spike sequences, where the accessible information is restricted to a local temporal neighborhood. Moreover, spike fluctuations hinder reliable intensity extraction. Both factors affect the performance of SCSR. To address these issues, we propose a hierarchical recurrent network named Spk2VidNet to reconstruct high-fidelity high resolution image sequences from low resolution spike data. To mitigate fluctuations, Spk2VidNet progressively exploits temporal correlations within spike stream to enhance feature representation by hierarchically enlarging temporal receptive fields. Within the recurrent block, we introduce an alignment module that leverages the motion consistency among multiple frames to jointly estimate and mutually refine inter-frame motions, achieving more accurate temporal alignment. In addition, we develop a fusion module to adaptively integrate neighboring aligned features based on multi-scale similarity for robust feature aggregation. We further propose a segment-wise training with state transfer strategy to efficiently model long-term dependencies with limited GPU memory. Experiments on synthetic and real-captured spike data demonstrate that Spk2VidNet achieves state-of-the-art performance.

Dynamic objects in our physical 4D (3D + time) world are constantly evolving, deforming, and interacting with other objects, leading to diverse 4D scene dynamics. In this paper, we present a universal generative pipeline, CHORD, for CHOReographing Dynamic objects and scenes and synthesizing this type of phenomena. Traditional rule-based graphics pipelines to create these dynamics are based on category-specific heuristics, yet are labor-intensive and not scalable. Recent learning-based methods typically demand large-scale datasets, which may not cover all object categories in interest. Our approach instead inherits the universality from the video generative models by proposing a distillation-based pipeline to extract the rich Lagrangian motion information hidden in the Eulerian representations of 2D videos. Our method is universal, versatile, and category-agnostic. We demonstrate its effectiveness by conducting experiments to generate a diverse range of multi-body 4D dynamics, show its advantage compared to existing methods, and demonstrate its applicability in generating robotics manipulation policies.

While real-world applications increasingly demand intricate scene manipulation, existing instruction-guided image editing benchmarks often oversimplify task complexity and lack comprehensive, fine-grained instructions. To bridge this gap, we introduce CompBench, a large-scale benchmark specifically designed for complex instruction-guided image editing. CompBench features challenging editing scenarios that incorporate fine-grained instruction following, spatial and contextual reasoning, thereby enabling comprehensive evaluation of image editing models' precise manipulation capabilities. To construct CompBench, we propose an MLLM-human collaborative framework with tailored task pipelines. Furthermore, we propose an instruction decoupling strategy that disentangles editing intents into four key dimensions: location, appearance, dynamics, and objects, ensuring closer alignment between instructions and complex editing requirements. Extensive evaluations reveal that CompBench exposes fundamental limitations of current image editing models and provides critical insights for the development of next-generation instruction-guided image editing systems. Our project page is available at https://comp-bench.github.io/.