CVPR 2026 论文列表

The incorporation of online reinforcement learning (RL) into diffusion and flow-based generative models has recently gained attention as a powerful paradigm for aligning model behavior with human preferences. By leveraging stochastic sampling via Stochastic Differential Equations (SDEs) during the denoising phase, these models can explore a variety of denoising trajectories, enhancing the exploratory capacity of RL. However, despite their ability to discover potentially high-reward samples, current approaches often struggle to effectively align with preferences due to the sparsity and narrowness of reward feedback. To overcome this limitation, we introduce a novel framework called Granular-GRPO (G^2RPO), which enables fine-grained and comprehensive evaluation of sampling directions in the RL training of flow models. Specifically, we propose a Singular Stochastic Sampling mechanism that supports step-wise stochastic exploration while ensuring strong correlation between injected noise and reward signals, enabling more accurate credit assignment to each SDE perturbation. Additionally, to mitigate the bias introduced by fixed-granularity denoising, we design a Multi-Granularity Advantage Integration module that aggregates advantages computed across multiple diffusion scales, resulting in a more robust and holistic assessment of sampling trajectories. Extensive experiments on various reward models, including both in-domain and out-of-domain settings, demonstrate that our G^2RPO outperforms existing flow-based GRPO baselines, highlighting its effectiveness and generalization capability.

Spiking Neural Networks (SNNs) have gained significant attention due to their event-driven computational paradigm, making them promising for neuromorphic computing. In recent years, the integration of SNNs and Transformer architectures has made remarkable progress in various tasks. However, existing spiking self-attention mechanisms predominantly focus on spatial information while neglecting explicit temporal modeling, leading to suboptimal performance. In this paper, we introduce the Temporal Interaction Coefficient (TIC) to analyze temporal dependency patterns in these spatial-only attention mechanisms, revealing their limited temporal interactions and restricted pattern diversity. To overcome this issue, we propose the Multi-Delay Mixer (MD-Mixer), drawing inspiration from time delay mechanisms in the nervous system. Specifically, MD-Mixer introduces multiple temporal delays to perform effective time mixing and facilitate temporally enriched spatial attention. In addition, it can be integrated seamlessly into existing Spiking Transformers as a drop-in replacement while maintaining computational efficiency. Extensive evaluations on static and neuromorphic benchmarks demonstrate that MD-Mixer substantially improves the performance of Spiking Transformers, outperforming existing state-of-the-art (SOTA) methods. This work establishes MD-Mixer as an effective and general solution for temporal modeling in event-driven architectures.

Statistically consistent methods based on the noise transition matrix (T) offer a theoretically grounded solution to Learning with Noisy Labels (LNL), with guarantees of convergence to the optimal clean-data classifier. In practice, however, these methods are often outperformed by empirical approaches such as sample selection, and this gap is usually attributed to the difficulty of accurately estimating T. The common assumption is that, given a perfect T, noise-correction methods would recover their theoretical advantage. In this work, we put this longstanding hypothesis to a decisive test. We conduct experiments under idealized conditions, providing correction methods with a perfect, oracle transition matrix. Even under these ideal conditions, we observe that these methods still suffer from performance collapse during training. This compellingly demonstrates that the failure is not fundamentally a T-estimation problem, but stems from a more deeply rooted flaw. To explain this behaviour, we provide a unified analysis that links three levels: macroscopic convergence states, microscopic optimisation dynamics, and information-theoretic limits on what can be learned from noisy labels. Together, these results give a formal account of why ideal noise correction fails and offer concrete guidance for designing more reliable methods for learning with noisy labels.

Recent significant advances in 3D scene representation have been driven by 3D Gaussian Splatting (3DGS), which has enabled real-time rendering with photorealistic quality. 3DGS often requires a large number of primitives to achieve high fidelity, leading to redundant representations and high resource consumption, thereby limiting its scalability for complex or large-scale scenes. Consequently, effective pruning strategies and more expressive primitives that can reduce redundancy while preserving visual quality are crucial for practical deployment. We propose an efficient, integrated reconstruction-aware pruning strategy that adaptively determines pruning timing and refining intervals based on reconstruction quality, thus reducing model size while enhancing rendering quality. Moreover, we introduce a 3D Difference-of-Gaussians primitive that jointly models both positive and negative densities in a single primitive, improving the expressiveness of Gaussians under compact configurations. Our method significantly improves model compactness, achieving up to 90% reduction in Gaussian-count while delivering visual quality that is similar to, or in some cases better than, that produced by state-of-the-art methods.

Multi-Label Recognition (MLR) based on Vision-Language Models (VLMs) aims to leverage their pre-trained knowledge to better adapt complex recognition scenarios, thereby enhancing model robustness. However, for realistic decentralized applications requiring federated learning, adapting VLMs to each client that possesses private and heterogeneous data can cause the model to overfit spurious label correlations, consequently triggering irrelevant categories when encountering new samples. To tackle this problem, we reconsider the federated learning for MLR with a causal model, in which we adopt a front-door adjustment and decouple the MLR modeling process by intermediate variables that magnify the oracle label co-occurrence. Guided by our analysis, we propose our FedMPT, the first method specifically designed for federated MLR. The core idea of FedMPT is to leverage generalizable conditions to steer federated MLR to mitigate erroneous label activations. To achieve this, FedMPT introduces an Large Language Model (LLM)-driven pipeline to decipher the underlying conditions that govern label dependencies. Furthermore, we introduce an optimal transport between the condition-enriched prompts and the image patches to uncover multiple region-level semantics. Finally, we generate synergistic predictions from different conditions with a crafted gating mechanism. Experiments on multiple benchmark datasets show that our proposed approach achieves competitive results and outperforms SOTA methods under varied settings.

Vision-language models (VLMs) have shown promise in earth observation (EO), yet they struggle with tasks that require grounding complex spatial reasoning in precise pixel-level visual representations. To address this problem, we introduce TerraScope, a unified VLM that delivers pixel-grounded geospatial reasoning with two key capabilities: (1) modality-flexible reasoning: it handles single-modality inputs (optical or SAR) and adaptively fuses different modalities into the reasoning process when both are available; (2) multi-temporal reasoning: it integrates temporal sequences for change analysis across multiple time points. In addition, we curate Terra-CoT, a large-scale dataset containing 1 million samples with pixel-level masks embedded in reasoning chains across multiple sources. We also propose TerraScope-Bench, the first benchmark for pixel-grounded geospatial reasoning with six sub-tasks that evaluates both answer accuracy and mask quality to ensure authentic pixel-grounded reasoning. Experiments show that TerraScope significantly outperforms existing VLMs on pixel-grounded geospatial reasoning while providing interpretable visual evidence.

Though recent advances in vision-language models (VLMs) have achieved remarkable progress across a wide range of multimodal tasks, understanding 3D spatial relationships from limited views remains a significant challenge. Previous reasoning methods typically rely on pure text (e.g., topological cognitive maps) or on 2D visual cues. However, their limited representational capacity hinders performance in specific tasks that require 3D spatial imagination. To address this limitation, we propose 3DThinker, a framework that can effectively exploit the rich geometric information embedded within images while reasoning, like humans do. Our framework is the first to enable 3D mentaling during reasoning without any 3D prior input, and it does not rely on explicitly labeled 3D data for training. Specifically, our training consists of two stages. First, we perform supervised training to align the 3D latent generated by VLM while reasoning with that of a 3D foundation model (e.g., VGGT). Then, we optimize the entire reasoning trajectory solely based on outcome signals, thereby refining the underlying 3D mentaling. Extensive experiments across multiple benchmarks show that 3DThinker consistently outperforms strong baselines and offers a new perspective toward unifying 3D representations into multimodal reasoning. Our code is available at https://github.com/zhangquanchen/3DThinker.

Omnimodal large language models (OmniLLMs) have attracted increasing research attention of late towards unified audio-video understanding. However, the high computational cost of processing longer joint audio-video token sequences has become a key bottleneck. Existing token compression methods have not addressed the emerging need to jointly compress multimodal tokens. To bridge this gap, we present OmniZip, a training-free, audio-guided audio-visual token-compression framework that optimizes multimodal token representation and accelerates model inference. Specifically, OmniZip first identifies salient audio tokens, then computes an audio retention score for each time group to capture information density, thereby dynamically guiding video token pruning and preserving cues from audio anchors enhanced by cross-modal similarity. For each time window, OmniZip compresses the video tokens using an interleaved spatio-temporal scheme. Extensive results demonstrate the merits of OmniZip: it achieves a 3.42x inference speedup and a 1.4x memory reduction over other top-performing counterparts, while maintaining the performance of OmniLLMs without training.

Recent advances in generative AI have dramatically improved photorealistic image synthesis, yet they fall short for studio-level multi-object compositing. This task demands simultaneous (i) near-perfect preservation of each item's identity, (ii) precise background and color fidelity, (iii) layout and design elements control, and (iv) complete, appealing displays showcasing all objects. However, current state-of-the-art models often alter object details, omit or duplicate objects, and produce layouts with incorrect relative sizing or inconsistent item presentations. To bridge this gap, we introduce PLACID, a framework that transforms a collection of object images into an appealing multi-object composite. Our approach makes two main contributions. First, we leverage a pretrained image-to-video (I2V) diffusion model with text control to preserve objects consistency, identities and background details by exploiting temporal priors from videos. Second, we propose a novel data curation strategy that generates synthetic sequences where randomly placed objects smoothly move to their target positions. This synthetic data aligns with the video model's temporal priors during training. At inference, objects initialized at random positions consistently converge into coherent layouts guided by text, with the final frame serving as the composite image. Extensive quantitative evaluations and user studies demonstrate that PLACID surpasses state-of-the-art methods in multi-object compositing, achieving superior identity, background, and color preservation, with less omitted objects and visually appealing results.

Diffusion Transformers (DiTs) set the state of the art in visual generation, yet their quadratic self-attention cost fundamentally limits scaling to long token sequences. Recent Top-K sparse attention approaches reduce the computation of DiTs by compressing tokens into block-wise representation and selecting a small set of relevant key blocks, but still suffer from (i) quadratic selection cost on compressed tokens and (ii) increasing K required to maintain model quality as sequences grow. We identify that their inefficiency is due to the single-level design, as a single coarse level is insufficient to represent the global structure.In this paper, we introduce Log-linear Sparse Attention (LLSA), a trainable sparse attention mechanism for extremely long token sequences that reduces both selection and attention costs from quadratic to log-linear complexity by utilizing a hierarchical structure. LLSA performs hierarchical Top-K selection, progressively adopting sparse Top-K selection with the indices found at the previous level,and introduces a Hierarchical KV Enrichment mechanism that preserves global context while using fewer tokens of different granularity during attention computation. To support efficient training, we develop a high-performance GPU implementation that uses only sparse indices for both the forward and backward passes, eliminating the need for dense attention masks.We evaluate LLSA on high-resolution pixel-space image generation without using patchification and VAE encoding. LLSA accelerates attention inference by 28.27 xand DiT training by 6.09 xon 256 x256 pixel token sequences, while maintaining generation quality. The results demonstrate that LLSA offers a promising direction for training long-sequence DiTs efficiently.

Remote photoplethysmography (rPPG) measurement enables non-contact physiological monitoring but suffers from accuracy degradation under head motion and illumination changes. Existing deep learning methods are mostly heuristic and lack theoretical grounding, limiting robustness and interpretability. In this work, we propose a physics-informed rPPG paradigm derived from the Navier-Stokes equations of hemodynamics, showing that the pulse signal follows a second-order dynamical system whose discrete solution naturally leads to a causal convolution, justifying the use of a Temporal Convolutional Network (TCN). Based on this principle, we design the PHASE-Net, a lightweight model with three key components: 1) Zero-FLOPs Axial Swapper module to swap or transpose a few spatial channels to mix distant facial regions, boosting cross-region feature interaction without changing temporal order; 2) Adaptive Spatial Filter to learn a soft spatial mask per frame to highlight signal-rich areas and suppress noise for cleaner feature maps; and 3) Gated TCN, a causal dilated TCN with gating that models long-range temporal dynamics for accurate pulse recovery. Extensive experiments demonstrate that PHASE-Net achieves state-of-the-art performance and strong efficiency, offering a theoretically grounded and deployment-ready rPPG solution.

Despite the remarkable success, recent reconstruction-based anomaly detection (AD) methods via diffusion modeling still involve fine-grained noise-strength tuning and computationally expensive multi-step denoising, leading to a fundamental tension between fidelity and efficiency. In this paper, we propose InvAD, a novel inversion-based anomaly detection approach for detection via noising in latent space, which circumvents explicit reconstruction. Importantly, we contend that the limitations of prior reconstruction-based methods originate from the prevailing "detection via denoising in RGB space" paradigm. To address this, we model AD under a reconstruction-free formulation that directly infers the final latent variable corresponding to the input image via DDIM inversion and then measures the deviation based on the known prior distribution for anomaly scoring. Specifically, when approximating the original probability flow ODE using the Euler method, we use only a few inversion steps to noise the clean image and improve inference efficiency. As the added noise is adaptively derived from the learned diffusion model, the original features of the clean test image can still be leveraged to yield high detection accuracy. We perform extensive experiments and detailed analysis across four widely used AD benchmarks under the unsupervised unified setting to demonstrate the effectiveness of our model, achieving state-of-the-art AD performance and about a twofold inference-time speedup without diffusion distillation.

Conventional 3D instance segmentation methods rely on labor-intensive 3D annotations for supervised training, which limits their scalability and generalization to novel objects. Recent approaches leverage multi-view 2D masks from the Segment Anything Model (SAM) to guide the merging of 3D geometric primitives, thereby enabling zero-shot 3D instance segmentation. However, these methods typically process each frame independently and rely solely on 2D metrics, such as SAM prediction scores, to produce segmentation maps. This design overlooks multi-view correlations and inherent 3D priors, leading to inconsistent 2D masks across views and ultimately fragmented 3D segmentation. In this paper, we propose MV3DIS, a coarse-to-fine framework for zero-shot 3D instance segmentation that explicitly incorporates 3D priors. Specifically, we introduce a 3D-guided mask matching strategy that uses coarse 3D segments as a common reference to match 2D masks across views and consolidates multi-view mask consistency via 3D coverage distributions. Guided by these view-consistent 2D masks, the coarse 3D segments are further refined into precise 3D instances. Additionally, we introduce a depth consistency weighting scheme that quantifies projection reliability to suppress ambiguities from inter-object occlusions, thereby improving the robustness of 3D-to-2D correspondence. Extensive experiments on the ScanNetV2, ScanNet200, ScanNet++, Replica, and Matterport3D datasets demonstrate the effectiveness of MV3DIS, which achieves superior performance over previous methods.

Architectural floor plan design demands joint reasoning over geometry, semantics, and spatial hierarchy, which remains a major challenge for current AI systems. Although recent diffusion and language models improve visual fidelity, they still struggle with coherent spatial reasoning and controllable generation. We present HouseMind, a multimodal large language model that unifies floor plan understanding, generation, and editing in one framework. We introduce discrete room-instance tokens to construct a unified vocabulary that bridges layouts and symbolic reasoning. With multimodal alignment and instruction tuning, the model synthesizes coherent, controllable layouts from text instructions. Experiments show how the framework achieves superior geometric validity and controllability while remaining efficient and locally deployable.

In this paper, we introduce NAS3R, a self-supervised feed-forward framework that jointly learns explicit 3D geometry and camera parameters with no ground-truth annotations and no pretrained priors. During training, NAS3R reconstructs 3D Gaussians from uncalibrated and unposed context views and renders target views using its self-predicted camera parameters, enabling self-supervised training from 2D photometric supervision. To ensure stable convergence, NAS3R integrates reconstruction and camera prediction within a shared transformer backbone regulated by masked attention, and adopts a depth-based Gaussian formulation that facilitates well-conditioned optimization. The framework is compatible with state-of-the-art supervised 3D reconstruction architectures and can incorporate pretrained priors or intrinsic information when available. Extensive experiments show that NAS3R achieves superior results to other self-supervised methods, establishing a scalable and geometry-aware paradigm for 3D reconstruction from unconstrained data. Code and models are publicly available at https://ranrhuang.github.io/nas3r/.

Understanding the physical world is essential for generalist AI agents. However, it remains unclear whether state-of-the-art vision perception models (e.g., large VLMs) can perform quantitative physical reasoning tasks. Existing evaluations are predominantly VQA-based and qualitative, offering limited insight into whether these models can infer the kinematic quantities of moving objects from videos. To address this, we present QuantiPhy, the first benchmark designed to quantitatively measure a VLM's physical reasoning ability. Comprising more than 3.3K video-text instances with numerical ground truth, QuantiPhy evaluates a VLM's performance on estimating an object's size, velocity, and acceleration at a given timestamp, using one of these properties as an input prior. The benchmark standardizes prompts and scoring to assess numerical accuracy, enabling fair comparisons across models. Our experiments on state-of-the-art VLMs reveal a consistent gap between their qualitative plausibility and actual numerical correctness. We further provide an in-depth analysis of key factors like background noise, counterfactual priors, and strategic prompting and find that state-of-the-art VLMs lean heavily on pre-trained world knowledge rather than faithfully using the provided visual and textual inputs when reasoning about objects' kinematic properties. QuantiPhy offers the first rigorous, scalable testbed to move VLMs beyond mere verbal plausibility toward a numerically grounded physical understanding.

We introduce MeshLAM, a feed-forward framework for one-shot animatable mesh head reconstruction that generates high-fidelity, animatable 3D head avatars from a single image. Unlike previous work that relies on time-consuming test-time optimization or extensive multi-view data, our method produces complete mesh representations with inherent animatability from a single image in a single forward pass. Our approach employs a dual shape and texture map architecture that simultaneously processes mesh vertices and texture map with extracted image features from a shared transformer backbone, allowing for coherent shape carving and appearance modeling. To prevent mesh collapse and ensure topological integrity during feed-forward deformation, we propose an iterative GRU-based decoding mechanism with progressive geometry deformation and texture refinement, coupled with a novel reprojection-based texture guidance mechanism that anchors appearance learning to the input image. Extensive experiments demonstrate that our method outperforms state-of-the-art approaches in reconstruction quality, animation capability, and computational efficiency. Project page at https://meshlam.github.io.

Recent advances in text-to-image (T2I) generation via reinforcement learning (RL) have benefited from reward models that assess semantic alignment and visual quality. However, most existing reward models pay limited attention to fine-grained spatial relationships, often producing images that appear plausible overall yet contain inaccuracies in object positioning. In this work, we present SpatialReward, a verifiable reward model explicitly designed to evaluate spatial layouts in generated images. SpatialReward adopts a multi-stage pipeline: a Prompt Decomposer extracts entities, attributes, and spatial metadata from free-form prompts; expert detectors provide accurate visual grounding of object positions and attributes; and a vision-language model applies chain-of-thought reasoning over grounded observations to assess complex spatial relations that are challenging for rule-based methods. To more comprehensively evaluate spatial relationships in generated images, we introduce SpatRelBench, a benchmark covering object attributes, orientation, inter-object relations, and rendered text placement. Experiments on Stable Diffusion and FLUX show that incorporating SpatialReward into RL training consistently improves spatial consistency and overall generation quality, with results aligned more closely to human judgments. These findings indicate that verifiable reward models hold considerable potential for enabling more accurate and controllable optimization in text-to-image generation models.

Recent advances in diffusion models have demonstrated impressive capability in generating high-quality images for simple prompts. However, when confronted with complex prompts involving multiple objects and hierarchical structures, existing models struggle to accurately follow instructions, leading to issues such as concept omission, confusion, and poor compositionality. To address these limitations, we propose a Hierarchical Compositional Generative framework (HiCoGen) built upon a novel Chain of Synthesis (CoS) paradigm. Instead of monolithic generation, HiCoGen first leverages a Large Language Model (LLM) to decompose complex prompts into minimal semantic units. It then synthesizes these units iteratively, where the image generated in each step provides crucial visual context for the next, ensuring all textual concepts are faithfully constructed into the final scene. To further optimize this process, we introduce a reinforcement learning (RL) framework. Crucially, we identify that the limited exploration of standard diffusion samplers hinders effective RL. We theoretically prove that sample diversity is maximized by concentrating stochasticity in the early generation stages and, based on this insight, propose a novel Decaying Stochasticity Schedule to enhance exploration. Our RL algorithm is then guided by a hierarchical reward mechanism that jointly evaluates the image at the global, subject, and relationship levels. We also construct HiCoPrompt, a new text-to-image benchmark with hierarchical prompts for rigorous evaluation. Experiments show our approach significantly outperforms existing methods in both concept coverage and compositional accuracy.