CVPR 2026 Papers

Smart factories use advanced technologies to optimize production and increase efficiency. To this end, the recognition of worker activity allows for accurate quantification of performance metrics, improving efficiency holistically while contributing to worker safety. OpenMarcie is, to the best of our knowledge, the biggest multimodal dataset designed for human action monitoring in manufacturing environments. It includes data from wearables sensing modalities and cameras distributed in the surroundings. The dataset is structured around two experimental settings, involving a total of 36 participants. In the first setting, twelve participants perform a bicycle assembly and disassembly task under semi-realistic conditions without a fixed protocol, promoting divergent and goal-oriented problem-solving. The second experiment involves twenty-five volunteers (24 valid data) engaged in a 3D printer assembly task, with the 3D printer manufacturer's instructions provided to guide the volunteers in acquiring procedural knowledge. This setting also includes sequential collaborative assembly, where participants assess and correct each other's progress, reflecting real-world manufacturing dynamics. OpenMarcie includes over 37 hours of egocentric and exocentric, multimodal, and multipositional data, featuring eight distinct data types and more than 200 independent information channels. The dataset is benchmarked across three human activity recognition tasks: activity classification, open vocabulary captioning, and cross-modal alignment.

First-Frame Propagation (FFP) offers a promising paradigm for controllable video editing, but existing methods are hampered by a reliance on cumbersome run-time guidance. We identify the root cause of this limitation as the inadequacy of current training datasets, which are often too short, low-resolution, and lack the task diversity required to teach robust temporal priors. To address this foundational data gap, we first introduce FFP-300K, a new large-scale dataset comprising 300K high-fidelity video pairs at 720p resolution and 81 frames in length, constructed via a principled two-track pipeline for diverse local and global edits. Building on this dataset, we propose a novel framework designed for true guidance-free FFP that resolves the critical tension between maintaining first-frame appearance and preserving source video motion. Architecturally, we introduce Adaptive Spatio-Temporal RoPE (AST-RoPE), which dynamically remaps positional encodings to disentangle appearance and motion references. At the objective level, we employ a self-distillation strategy where an identity propagation task acts as a powerful regularizer, ensuring long-term temporal stability and preventing semantic drift. Comprehensive experiments on the EditVerseBench benchmark demonstrate that our method significantly outperforming existing academic and commercial models by receiving about 0.2 PickScore and 0.3 VLM score improvement against these competitors. The project is at ffp-300k.github.io.

Floorplan localization aims to estimate the camera pose of a query image with respect to a 2D floorplan, providing a lightweight and long-term stable alternative to localization based on 3D maps or large image databases for indoor robotics and AR. Recent methods frame the problem as ray-based matching, representing the image as a set of rays annotated with depth or semantic labels and aligning them with the floorplan. However, they still face challenges in addressing the complexity of indoor environments, which can be decomposed into environmental, geometric, and semantic ambiguities.To address these ambiguities, we propose a floorplan-aware probabilistic fusion framework that models both depth and semantic information within a unified architecture. Our framework also combines a distribution-based ray confidence estimator, which down-weights uncertain geometric hypotheses, with a probabilistic semantic matching scheme based on Jensen-Shannon divergence (JSD), which preserves and leverages informative semantic ambiguity instead of collapsing it into hard labels. Experiments on challenging benchmarks demonstrate that our approach significantly outperforms prior methods in both robustness and accuracy.

We present PhysInOne, a large-scale synthetic dataset addressing the critical scarcity of physically-grounded training data for AI systems. Unlike existing datasets limited to merely hundreds or thousands of examples, PhysInOne provides 2 million videos across 153,810 dynamic 3D scenes, covering 71 basic physical phenomena in mechanics, optics, fluid dynamics, and magnetism. Distinct from previous works, our scenes feature multiobject interactions against complex backgrounds, with comprehensive ground-truth annotations including 3D geometry, semantics, dynamic motion, physical properties, and text descriptions. We demonstrate PhysInOne's efficacy across four emerging applications: physics-aware video generation, long-/short-term future frame prediction, physical property estimation, and motion transfer. Experiments show that fine-tuning foundation models on PhysInOne significantly enhances physical plausibility, while also exposing critical gaps in modeling complex physical dynamics and estimating intrinsic properties. As the largest dataset of its kind, orders of magnitude beyond prior works, PhysInOne establishes a new benchmark for advancing physics-grounded world models in generation, simulation, and embodied AI.

Machine learning has been progressively generalised to operate within non-Euclidean domains, but geometrically accurate methods for learning on surfaces are still falling behind. The lack of closed-form Riemannian operators, the non-differentiability of their discrete counterparts, and poor parallelisation capabilities have been the main obstacles to the development of the field on meshes. A principled framework to compute the exponential map on Riemannian surfaces discretised as meshes is straightest geodesics, which also allows to trace geodesics and parallel-transport vectors as a by-product. We provide a parallel GPU implementation and derive two different methods for differentiating through the straightest geodesics, one leveraging an extrinsic proxy function and one based upon a geodesic finite differences scheme. After proving our parallelisation performance and accuracy, we demonstrate how our differentiable exponential map can supercharge geometrically-correct learning and optimisation pipelines. In particular, to showcase the versatility of our method, we propose a new geodesic convolutional layer, a new flow matching method for learning on meshes, and a second-order optimiser that we apply to centroidal Voroni tesselation. Our code, pre-trained models, and pip-installable library will be made available upon publication.

Polarization-aware Neural Radiance Fields (NeRF) enables novel view synthesis of specular scenes but suffers from slow training, inefficient rendering, and material/viewpoint assumptions. 3D Gaussian Splatting (3DGS) supports real-time rendering but struggles with reflection reconstruction due to reflection-geometry entanglement. We propose PolarGuide-GSDR, a polarization-guided framework that leverages polarization information to separate specular reflections for guided reconstruction, and builds bidirectional iterative optimization between polarization and 3DGS: it first uses 3DGS geometric priors to disambiguate polarization, then uses refined polarization normal to guide 3DGS normals and Spherical Harmonics representations, achieving high-fidelity reflection separation and full-scene reconstruction without restrictive material assumptions. Experiments on public and self-collected datasets show that PolarGuide-GSDR achieves state-of-the-art performance in specular reconstruction, normal estimation, and novel view synthesis, while maintaining real-time rendering. To our knowledge, this is the first framework embedding polarization priors into 3DGS optimization, offering superior interpretability and real-time performance for complex reflective scenes.

Fine-tuning approaches for Vision-Language Models (VLMs) face a critical three-way trade-off between In-Distribution (ID) accuracy, Out-of-Distribution (OOD) generalization, and adversarial robustness. Existing robust fine-tuning strategies resolve at most two axes of this trade-off. Generalization-preserving methods retain ID/OOD performance but leave models vulnerable to adversarial attacks, while adversarial training improves robustness to targeted attacks but degrades ID/OOD accuracy. Our key insight is that the robustness trade-off stems from two geometric failures: sharp, anisotropic minima in parameter space and unstable feature representations that deform under perturbation. To address this, we propose GRACE (Gram-aligned Robustness via Adaptive Curvature Estimation), a unified fine-tuning framework that jointly regularizes the parameter-space curvature and feature-space invariance for VLMs. Grounded in Robust PAC-Bayes theory, GRACE employs adaptive weight perturbations scaled by local curvature to promote flatter minima, combined with a feature alignment loss that maintains representation consistency across clean, adversarial, and OOD inputs. On ImageNet fine-tuning of CLIP models, GRACE simultaneously improves ID accuracy by 10.8%, and adversarial accuracy by 13.5% while maintaining 57.0% OOD accuracy (vs. 57.4% zero-shot baseline). Geometric analysis confirms that GRACE converges to flatter minima without feature distortion across distribution shifts, providing a principled step toward generalized robustness in foundation VLMs.

Recent advances in audio-visual representation learning have shown the value of combining contrastive alignment with masked reconstruction. However, jointly optimizing these objectives in a single forward pass forces the contrastive branch to rely on randomly visible patches designed for reconstruction rather than cross-modal alignment, introducing semantic noise and optimization interference. We propose TG-DP, a Teacher-Guided Dual-Path framework that decouples reconstruction and alignment into separate optimization paths. By disentangling the masking regimes of the two branches, TG-DP enables the contrastive pathway to use a visibility pattern better suited to cross-modal alignment. A teacher model further provides auxiliary guidance for organizing visible tokens in this branch, helping reduce interference and stabilize cross-modal representation learning. TG-DP achieves state-of-the-art performance in zero-shot retrieval. On AudioSet, it improves R@1 from 35.2% to 37.4% for video-to-audio retrieval and from 27.9% to 37.1% for audio-to-video retrieval. The learned representations also remain semantically robust, achieving state-of-the-art linear-probe performance on AS20K and VGGSound. Taken together, our results suggest that decoupling multimodal objectives and introducing teacher-guided structure into the contrastive pathway provide an effective framework for improving large-scale audio-visual pretraining.

Conventional evaluation methods for multimodal LLMs (MLLMs) lack interpretability and are often insufficient to fully disclose significant capability gaps across models. To address this, we introduce AuditDM, an automated framework that actively discovers and rectifies MLLM failure modes by auditing their divergence. AuditDM fine-tunes an MLLM as an auditor via reinforcement learning to generate challenging questions and counterfactual images that maximize disagreement among target models. Once trained, the auditor uncovers diverse, interpretable exemplars that reveal model weaknesses and serve as annotation-free data for rectification. When applied to SoTA models like Gemma-3 and PaliGemma-2, AuditDM discovers more than 20 distinct failure types. Fine-tuning on these discoveries consistently improves all models across 16 benchmarks, and enables a 3B model to surpass its 28B counterpart. Our results suggest that as data scaling hits diminishing returns, targeted model auditing offers an effective path to model diagnosis and improvement.

3D reconstruction from multi-view images is a core challenge in computer vision. Recently, feed-forward methods have emerged as efficient and robust alternatives to traditional per-scene optimization techniques. Among them, state-of-the-art models like the Visual Geometry Grounding Transformer (VGGT) leverage full self-attention over all image tokens to capture global relationships. However, this approach suffers from poor scalability due to the quadratic complexity of self-attention and the large number of tokens generated in long image sequences.In this work, we introduce FlashVGGT, an efficient alternative that addresses this bottleneck through a descriptor-based attention mechanism. Instead of applying dense global attention across all tokens, FlashVGGT compresses spatial information from each frame into a compact set of **descriptor tokens**. Global attention is then computed as cross-attention between the full set of image tokens and this smaller descriptor set, significantly reducing computational overhead.Moreover, the compactness of the descriptors enables online inference over long sequences via a chunk-recursive mechanism that reuses cached descriptors from previous chunks. Experimental results show that FlashVGGT achieves reconstruction accuracy competitive with VGGT while reducing inference time to just **9.3%** for 1,000 images, and scaling efficiently to sequences exceeding **3,000** images.

Vision-Language-Action (VLA) models excel in robotic manipulation but are constrained by their heavy reliance on expert demonstrations, leading to demonstration bias and limiting performance. Reinforcement learning (RL) is a vital post-training strategy to overcome these limits, yet current VLA-RL methods, including group-based optimization approaches, are crippled by severe reward sparsity. Relying on binary success indicators wastes valuable information in failed trajectories, resulting in low training efficiency. To solve this, we propose Self-Referential Policy Optimization (SRPO), a novel VLA-RL framework. SRPO eliminates the need for external demonstrations or manual reward engineering by leveraging the model's own successful trajectories, generated within the current training batch, as a self-reference. This allows us to assign a progress-wise reward to failed attempts. A core innovation is the use of Latent World Representations to measure behavioral progress robustly. Instead of relying on raw pixels or requiring domain-specific fine-tuning, we utilize the compressed, transferable encodings from a world model's latent space. These representations naturally capture progress patterns across environments, enabling accurate, generalized trajectory comparison. Empirical evaluations on the LIBERO benchmark demonstrate SRPO's efficiency and effectiveness. Starting from a supervised baseline with 48.9% success, SRPO achieves a new state-of-the-art success rate of 99.2% in just 200 RL steps, representing a 103% relative improvement without any extra supervision. Furthermore, SRPO shows substantial robustness, achieving a 167% performance improvement on the LIBERO-Plus benchmark.

Recent progress in large language models (LLMs) has shown that reasoning improves when intermediate thoughts are externalized into explicit workspaces, such as chain-of-thought traces or tool-augmented reasoning. Yet, visual language models (VLMs) lack an analogous mechanism for spatial reasoning, limiting their ability to generate images that accurately reflect geometric relations, object identities, and compositional intent. We introduce the concept of a spatial scratchpad -- a 3D reasoning substrate that bridges linguistic intent and image synthesis. Given a text prompt, our framework parses subjects and background elements, instantiates them as editable 3D meshes, and employs agentic scene planning for placement, orientation, and viewpoint selection. The resulting 3D arrangement is rendered back into the image domain with identity-preserving cues, enabling the VLM to generate spatially consistent and visually coherent outputs. Unlike prior 2D layout-based methods, our approach supports intuitive 3D edits that propagate reliably into final images. Empirically, it achieves a 32% improvement in text alignment on GenAI-Bench, demonstrating the benefit of explicit 3D reasoning for precise, controllable image generation. Our results highlight a new paradigm for vision-language models that deliberate not only in language, but also in space. Code and visualizations at https://oindrilasaha.github.io/3DScratchpad/

Vision Transformers (ViTs) often degrade under distribution shifts because they rely on spurious correlations, such as background cues, rather than semantically meaningful features. Existing regularization methods, typically relying on simple foreground-background masks, which fail to capture the fine-grained semantic concepts that define an object (e.g., "long beak" and "wings" for a "bird"). As a result, these methods provide limited robustness to distribution shifts. To address this limitation, we introduce a novel finetuning framework that steers model reasoning toward concept-level semantics. Our approach optimizes the model's internal relevance maps to align with spatially grounded concept masks. These masks are generated automatically, without manual annotation: class-relevant concepts are first proposed using an LLM-based, label-free method, and then segmented using a VLM. The finetuning objective aligns relevance with these concept regions while simultaneously suppressing focus on spurious background areas. Notably, this process requires only a minimal set of images and uses half of the dataset classes. Extensive experiments on five out-of-distribution benchmarks demonstrate that our method improves robustness across multiple ViT-based models. Furthermore, we show that the resulting relevance maps exhibit stronger alignment with semantic object parts, offering a scalable path toward more robust and interpretable vision models. Finally, we confirm that concept-guided masks provide more effective supervision for model robustness than conventional segmentation maps, supporting our central hypothesis.

Large vision-language models (VLMs) show strong multimodal understanding but still struggle with 3D spatial reasoning, such as distance estimation, size comparison, and cross-view consistency. Existing 3D-aware methods either depend on auxiliary 3D information or enhance RGB-only VLMs with geometry encoders through shallow feature fusion. We propose SpaceMind, a multimodal large language model trained on SpaceMind-700K for spatial reasoning from RGB inputs. The model adopts a dual-encoder architecture integrating VGGT as a spatial understanding encoder and InternViT as a 2D visual encoder. The key idea is to treat the camera representation as an active guiding modality rather than passive metadata. Specifically, SpaceMind introduces a lightweight Camera-Guided Modality Fusion module before the language model to replace shallow fusion. It applies camera-conditioned biasing to spatial tokens, assigns query-independent weights reflecting their geometric importance, and uses the camera embedding to gate the fused representation. Empirically, SpaceMind establishes new state-of-the-art results on VSI-Bench, SQA3D and SPBench, surpassing both open and proprietary systems on VSI-Bench and SPBench by large margins and achieving state-of-the-art performance on SQA3D. These results demonstrate that camera-guided modality fusion is an effective and practical inductive bias for equipping VLMs with genuinely spatially grounded intelligence. We will release code and model checkpoints to support future research.

Adversarial attacks have evolved from simply disrupting predictions on conventional task-specific models to the more complex goal of manipulating image semantics in Large Vision-Language Models (LVLMs). However, existing methods struggle with controllability and cannot precisely manipulate the semantics of specific concepts in an image. We attribute this limitation to semantic entanglement in the patch-token representations that adversarial attacks typically operate on: global context aggregated by self-attention in the vision encoder dominates patch features, making them unreliable for precise local semantic manipulation. Our systematic investigation reveals a key insight: value features, computed within the transformer attention block, provide much more precise handles for manipulation. We show that these value features suppress global-context channels, allowing them to retain high-entropy, disentangled local semantic information. Building on this discovery, we propose V-Attack, a novel method for precise local semantic attacks. V-Attack targets value features and introduces two core components: (1) a Self-Value Enhancement module to refine the intrinsic semantic richness of value features, and (2) a Text-Guided Value Manipulation module that uses text prompts to locate the source concept and optimize it toward a target concept. By bypassing entangled patch features, V-Attack achieves highly effective semantic control. Extensive experiments across diverse LVLMs, including LLaVA, InternVL, DeepSeek-VL, and GPT-4o, show that V-Attack improves attack success rate by an average of 36% over state-of-the-art methods, exposing critical vulnerabilities in modern vision-language understanding.

We present UniSH, a unified, feed-forward framework for joint metric-scale 3D scene and human reconstruction. A key challenge in this domain is the scarcity of large-scale, annotated real-world data, forcing a reliance on synthetic datasets. This reliance introduces a significant sim-to-real domain gap, leading to poor generalization, low-fidelity human geometry, and poor alignment on in-the-wild videos. To address this, we propose an innovative training paradigm that effectively leverages unlabeled in-the-wild data. Our framework bridges strong, disparate priors from scene reconstruction and HMR, and is trained with two core components: (1) a robust distillation strategy to refine human surface details by distilling high-frequency details from an expert depth model, and (2) a two-stage supervision scheme, which first learns coarse localization on synthetic data, then fine-tunes on real data by directly optimizing the geometric correspondence between the SMPL mesh and the human point cloud. This approach enables our feed-forward model to jointly recover high-fidelity scene geometry, human point clouds, camera parameters, and coherent, metric-scale SMPL bodies, all in a single forward pass. Extensive experiments demonstrate that our model achieves state-of-the-art performance on human-centric scene reconstruction and delivers highly competitive results on global human motion estimation, comparing favorably against both optimization-based frameworks and HMR-only methods. Project page: https://murphylmf.github.io/UniSH/.

Generative video-to-audio (V2A) models produce highly plausible soundtracks, but it remains unclear whether they capture the underlying physical processes. Existing evaluations emphasize perceptual realism and overlook physical correctness under controlled interventions. In this paper, we introduce FlatSounds, a benchmark that audits the physical reasoning of V2A models through: 1) controlled counterfactual pairs in which a single physical factor is varied, and 2) single-video pattern tests that probe internal consistency and directional trends. These settings test whether the generated audio correctly reflects specific physical properties and timings. Our evaluation of state-of-the-art models reveals a consistent trade-off: models rely more on text captions than the visual stream to infer physics and semantics. Captions generally improve physical and semantic accuracy, but paradoxically degrade temporal alignment. Our results highlight the need to move beyond audio quality toward learning physical processes directly from pixels. Finally, we find that our physics-based metrics correlate strongly with human preference tests on our own data.

While diffusion models have shown great potential in portrait generation, generating expressive, coherent, and controllable cinematic portrait videos remains a significant challenge. Existing intermediate signals for portrait generation, such as 2D landmarks and parametric models, have limited disentanglement capabilities and cannot express personalized details due to their sparse or low-rank representation. Therefore, existing methods based on these models struggle to accurately preserve subject identity and expressions, hindering the generation of highly expressive portrait videos. To overcome these limitations, we propose a high-fidelity personalized head representation that more effectively disentangles expression and identity. This representation captures both static, subject-specific global geometry and dynamic, expression-related details. Furthermore, we introduce an expression transfer module to achieve personalized transfer of head pose and expression details between different identities. We use this sophisticated and highly expressive head model as a conditional signal to train a diffusion transformer (DiT)-based generator to synthesize richly detailed portrait videos. Extensive experiments on self- and cross-reenactment tasks demonstrate that our method outperforms previous models in terms of identity preservation, expression accuracy, and temporal stability, particularly in capturing fine-grained details of complex motion.

Streaming video understanding often involves time-sensitive scenarios where models need to answer exactly when the supporting visual evidence appears: answering before the evidence reflects speculation, answering after it has passed reduces real-time utility. To capture this behavior, we introduce a readiness-aware formulation of streaming video understanding with the **Answer Readiness Score (ARS)**, a timing-aware objective with asymmetric early and late penalties. When combined with correctness, ARS defines an effective accuracy that measures not just whether a model is right, but whether it answers at the appropriate moment. Building on this formulation, we introduce **StreamReady**, a framework to unify temporal reasoning with on-time answering through a lightweight readiness mechanism that decides if sufficient evidence has been observed before responding. To evaluate this capability, we further introduce **ProReady-QA**, a benchmark with annotated answer evidence windows and proactive multi-turn questions across local and global contexts. StreamReady achieves superior performance on ProReady-QA, and consistently outperforms prior methods across eight additional streaming and offline long-video benchmarks, demonstrating robust and broadly generalizable video understanding capability.