CVPR 2026 论文列表

Generating 3D shapes at part level is pivotal for downstream applications such as mesh retopology, UV mapping, and 3D printing. However, existing part-based generation methods often lack sufficient controllability and suffer from poor semantically meaningful decomposition. To this end, we introduce X-Part, a controllable generative model designed to decompose a holistic 3D object into semantically meaningful and structurally coherent parts with high geometric fidelity. X-Part exploits the bounding box as prompts for the part generation and injects point-wise semantic features for meaningful decomposition. Furthermore, we design an editable pipeline for interactive part generation. Extensive experimental results show that X-Part achieves state-of-the-art performance in part-level shape generation. This work establishes a new paradigm for creating production-ready, editable, and structurally sound 3D assets. Codes will be released for public research.

Recent video editing models have achieved impressive results, but most still require large-scale paired datasets. Collecting such naturally aligned pairs at scale remains highly challenging and constitutes a critical bottleneck, especially for local video editing data. Existing workarounds transfer image editing to video through global motion control for pair-free video editing, but such designs struggle with background and temporal consistency. In this paper, we propose NOVA: Sparse Control & Dense Synthesis, a new framework for video editing. Specifically, the sparse branch provides semantic guidance through user-edited keyframes distributed across the video, and the dense branch continuously incorporates motion and texture information from the original video to maintain high fidelity and coherence. Moreover, we introduce a degradation-simulation training strategy that enables the model to learn motion reconstruction and temporal consistency by training on artificially degraded videos, thus eliminating the need for paired data. Our extensive experiments demonstrate that NOVA outperforms existing approaches in edit fidelity, motion preservation, and temporal coherence.

Dataset Distillation (DD) aims to compress large-scale datasets into a small number of condensed Images Per Class (IPC), enabling efficient network training. Previous coreset selection and synthetic-based DD methods achieve reasonable performance. However, our in-depth investigation reveals that existing methods share a common issue: pattern imbalance. Specifically, they either overemphasize class-general patterns representing the majority of each class or focus on fewer marginal patterns critical for model generalization. To address this issue, we propose a novel framework, Balanced Patterns Selection (BPS). Unlike prior methods that assume each class forms a single cluster, BPS models the multiple visual pattern distribution within each class via a hierarchical semantic structure inherent to the dataset. It then selects two complementary subsets in a balanced manner from the center (class-general patterns) and the margins (marginal patterns) of each pattern, producing a pattern-balanced coreset. Theoretically, we prove that the BPS-selected coreset aligns with the original dataset in both information coverage and performance. Moreover, its model-agnostic selection nature ensures cross-architecture generalization, while the Optimize-Once-for-All-IPCs property guarantees efficiency. Extensive experiments on four benchmarks demonstrate that BPS significantly outperforms existing state-of-the-art methods. The source code is available at: https://github.com/BeCarefulOfYournaoke/BPS.

Interpreting individual neurons or directions in activation space is an important topic in mechanistic interpretability. Numerous automated interpretability methods have been proposed to generate such explanations, but it remains unclear how reliable these explanations are, and which methods produce the most accurate descriptions. While crowd-sourced evaluations are commonly used, existing pipelines are noisy, costly, and typically assess only the highest-activating inputs, leading to unreliable results. In this paper, we introduce two techniques to enable cost-effective and accurate crowdsourced evaluation of automated interpretability methods beyond top activating inputs. First, we propose Model-Guided Importance Sampling (MG-IS) to select the most informative inputs to show human raters. In our experiments, we show this reduces the number of inputs needed to reach the same evaluation accuracy by 13x. Second, we address label noise in crowd-sourced ratings through Bayesian Rating Aggregation (BRAgg), which allows us to reduce the number of ratings per input required to overcome noise by 3x. Together, these techniques reduce the evaluation cost by 40x, making large-scale evaluation feasible. Finally, we use our methods to conduct a large scale crowd-sourced study comparing recent automated interpretability methods for vision networks.

Visual autoregressive (AR) generation models have demonstrated strong potential for image generation, yet their next-token-prediction paradigm introduces considerable inference latency. Although speculative decoding (SD) has been proven effective for accelerating visual AR models, its "draft one step, then verify one step" paradigm prevents a direct reduction in the number of forward passes, limiting its acceleration potential. Motivated by the interchangeability of visual tokens, we explore verification skipping in the SD process for the first time to explicitly cut the number of target model forward passes, thereby reducing inference latency. By analyzing the characteristics of the drafting stage, we observe that verification redundancy and stale feature reusability are key factors to maintain generation quality while improving speed for verification-free steps. Inspired by these two observations, we propose a novel SD framework VVS to accelerate \underline \text v isual AR model via partial \underline \text v erification \underline \text s kipping, which integrates three complementary modules: (1) a verification-free token selector with dynamically truncation, (2) token level feature caching and reuse, and (3) fine-grained skipped step scheduling. Consequently, VVS reduces the number of target model forward passes by 2.8x relative to vanilla AR decoding while maintaining competitive generation quality, offering a superior speed-quality trade-off over conventional SD frameworks and revealing strong potential to reshape the SD paradigm. Our code is available at https://github.com/HyattDD/VVS.

Stereo vision is fundamental for enabling machines to perceive and interact with the world. While monocular stereo methods offer hardware compactness, they struggle with generalization due to reliance on data-driven priors. Binocular and multi-view systems improve accuracy but incur higher hardware complexity and data inefficiency. In this paper, we introduce a monocular solution for high-framerate stereo vision via temporal optical modulation. The modulation directs light from two views onto a single sensor in a mixed manner, while periodically attenuating one view at 60 Hz. To capture the temporal variations introduced by this modulation, we employ a high-speed spike camera that records the mixed scene as temporally dense spikes. The high temporal resolution of these spikes enables the construction of a linear system for efficient binocular video decoupling. Consequently, we introduce a two-stage decoding methodology for achieving high-quality stereo vision: An efficient least-squares-based baseline reconstruction followed by a deep learning refinement module. Experimental results demonstrate that our approach achieves 240FPS binocular video reconstruction with superior accuracy compared to monocular systems, while maintaining the hardware compactness and data efficiency. Code is available at https://github.com/yongqiye00/MonoSpikeStereo.

The creation of 3D human avatars from multi-view videos is a significant yet challenging task in computer vision. However, existing techniques rely on high-quality, sharp images as input, which are often impractical to obtain in real-world scenarios due to variations in human motion speed and intensity. This paper introduces a novel method for directly reconstructing sharp 3D human Gaussian avatars from blurry videos. The proposed approach incorporates a 3D-aware, physics-based model of blur formation caused by human motion, together with a 3D human motion model designed to resolve ambiguities in motion-induced blur. This framework enables the joint optimization of the avatar representation and motion parameters from a coarse initialization. Comprehensive benchmarks are established using both a synthetic dataset and a real-world dataset captured with a 360-degree synchronous hybrid-exposure camera system. Extensive evaluations demonstrate the effectiveness of the model across diverse conditions. Codes Available: https://github.com/MyNiuuu/MAD-Avatar

This paper revisits camera pose estimation through the lens of self-supervised pretraining, focusing on inverse-dynamics pretraining as a scalable alternative to the current trend of fully supervised training with 3D annotations. Concretely, we employ inverse- and forward-dynamics models to learn latent action representations, similar to Genie from large-scale driving videos.Our idea is simple yet effective. Existing methods use latent actions in their original capacity, that is, as action conditioning of world-models or as proxies of robot action parameters in policy networks.Our method, dubbed LA-Pose, repurposes the latent action features as inputs to a camera pose estimator, finetuned on a limited set of high-quality 3D annotations.This formulation enables accurate and generalizable pose prediction while maintaining feed-forward efficiency. Extensive experiments on driving benchmarks show that LA-Pose achieves competitive and even superior performance to state-of-the-art methods while using orders of magnitude less labeled data. Concretely, on the Waymo and PandaSet benchmarks, LA-Pose achieves over 10% higher pose accuracy than recent feed-forward methods.To our knowledge, this work is the first to demonstrate the power of inverse-dynamics self-supervised learning for pose estimation.

Existing generative models, such as diffusion and auto-regressive networks, are inherently static, relying on a fixed set of pretrained parameters to handle all inputs. In contrast, humans flexibly adapt their internal generative representations to each perceptual or imaginative context. Inspired by this capability, we introduce Composer, a new paradigm for adaptive generative modeling based on test-time instance-specific parameter composition. Composer generates input-conditioned parameter adaptations at inference time, which are injected into the pretrained model's weights, enabling per-input specialization without fine-tuning or retraining. Adaptation occurs once prior to multi-step generation, yielding higher-quality, context-aware outputs with minimal computational and memory overhead. Experiments show that Composer substantially improves performance across diverse generative models and use cases, including lightweight/quantized models and test-time scaling. By leveraging input-aware parameter composition, Composer establishes a new paradigm for designing generative models that dynamically adapt to each input, moving beyond static parameterization. The code will be available at https://github.com/tmtuan1307/Composer.

Long-form video understanding remains challenging due to the extended temporal structure and dense multimodal cues. Despite recent progress, many existing approaches still rely on hand-crafted reasoning pipelines or employ token-consuming video preprocessing to guide MLLMs in autonomous reasoning. To overcome these limitations, we introduce VideoARM, an agentic reasoning-over-hierarchical-memory paradigm for long-form video understanding. Instead of static, exhaustive preprocessing, VideoARM performs adaptive, on-the-fly agentic reasoning and memory construction. Specifically, VideoARM performs an adaptive and continuous loop of observing, thinking, acting, and memorizing, where a controller autonomously invokes tools to interpret the video in a coarse-to-fine manner, thereby substantially reducing token consumption. In parallel, a hierarchical multimodal memory continuously captures and updates multi-level clues throughout the operation of the agent, providing precise contextual information to support the controller in decision-making. Experiments on prevalent benchmarks demonstrate that VideoARM outperforms the state-of-the-art method, DVD, while significantly reducing token consumption for long-form videos.

Performance estimation under distribution shift aims to predict how a model behaves on an unlabeled test set whose distribution differs from the training data, a scenario that requires reliable indicators that can faithfully reflect model behavior without ground-truth labels. Existing approaches rely solely on the outputs of the given model whose biases are amplified once the distribution shifts, weakening the correlation with the true performance. Motivated by this limitation, we propose Fused Reference Alignment Prediction (FRAP), which leverages the complementary strengths of an external foundation model and the base model to construct a more reliable surrogate of the ground-truth labels. FRAP aligns the prediction distribution of the foundation model with that of the base model by applying temperature-scaled calibration that minimizes their divergence. The aligned predictions are fused through confidence-based weighting into a refined reference distribution that integrates robustness from the foundation model and domain-specific expertise from the base model, and performance estimation is obtained by measuring how closely the base model predictions agree with this reference. Extensive experiments across diverse datasets and architectures show that FRAP provides consistent and substantial improvements over representative performance-estimation methods under distribution shift.

While multimodal large language models (MLLMs) have achieved remarkable success in recent advancements, their susceptibility to jailbreak attacks has come to light. In such attacks, adversaries exploit carefully crafted prompts to coerce models into generating harmful or undesirable content. Existing defense mechanisms often rely on external inference steps or safety alignment training, both of which are less effective and impractical when facing sophisticated adversarial perturbations in white-box scenarios. To address these challenges and bolster MLLM robustness, we introduce SAFEMLLM by adopting an adversarial training framework that alternates between an attack step for generating adversarial noise and a model updating step. At the attack step, SAFEMLLM generates adversarial perturbations through a newly proposed contrastive embedding attack (CoE-Attack), which optimizes token embeddings under a contrastive objective. SAFEMLLM then updates model parameters to neutralize the perturbation effects while preserving model utility on benign inputs. We evaluate SAFEMLLM across multiple MLLMs and six jailbreak methods spanning multiple modalities. Experimental results show that SAFEMLLM effectively defends against diverse attacks, maintaining robust performance and utilities.

We present FlashLips, a two-stage, mask-free lip-sync system that decouples lips control from rendering and achieves real-time performance, with our U-Net variant running at over 100 FPS on a single GPU, while matching the visual quality of larger state-of-the-art models. Stage 1 is a compact, one-step latent-space editor that reconstructs an image using a reference identity, a masked target frame, and a low-dimensional lips-pose vector, trained purely with reconstruction losses - no GANs or diffusion. To remove explicit masks at inference, we use self-supervision via mouth-altered target variants as pseudo ground truth, teaching the network to localize lip edits while preserving the rest. Stage 2 is an audio-to-pose transformer trained with a flow-matching objective to predict lips-pose vectors from speech. Together, these stages form a simple and stable pipeline that combines deterministic reconstruction with robust audio control, delivering high perceptual quality and faster-than-real-time speed.

Out-of-distribution (OOD) detection is essential for the safe deployment of machine learning models. While extensive work has focused on designing effective scoring functions for OOD detection, relatively few studies explore training neural networks with calibration-oriented objectives, which often compromise predictive accuracy and restrict the choice of scoring functions. In this work, we first identify feature collapse in Logit Normalization (LogitNorm), and then propose a novel hyperparameter-free training formulation that significantly improves a wide range of post-hoc detection methods. Specifically, we introduce a feature distance-aware normalization objective, termed ELogitNorm, which enhances both OOD detection performance and in-distribution (ID) confidence calibration. Extensive experiments on standard benchmarks demonstrate that our approach outperforms state-of-the-art training-time methods in OOD detection while preserving strong ID classification performance. Our code is available at: https://github.com/limchaos/ElogitNorm.

Visual models are deployed on many Internet-of-Things (IoT) devices to power a variety of visual applications at the network edge. These models often need to be fine-tuned on-device continually to adapt to changing operating environments timely. However, the computing and energy overheads incurred are often overwhelming for resource-constrained IoT devices. Existing methods score sample importance based on the entire model via multi-epoch training, incurring overhead that may even exceed the training overhead reduction. To reduce these fine-tuning overheads, this paper presents LoPrune, a novel data pruning method that identifies and removes samples with negligible contributions to model adaptation. The key idea is to evaluate sample importance via a Trainable Subspace Alignment (TSA) Score to align the importance estimation with accurate update directions of the learnable adapter, i.e., Low-Rank Adaptation (LoRA). Specifically, LoPrune projects the influence function onto the LoRA subspace, enforcing consistency between the importance score and the model's updatable directions while substantially reducing the problem's dimensionality. It then leverages Kronecker-Factored Approximate Curvature to approximate the change of learnable adapter induced by a sample as its TSA score, retaining higher-scoring samples. Experiments with four representative visual models fine-tuned on three datasets demonstrate that compared with the best state-of-the-art data pruning baselines, LoPrune can reduce fine-tuning overhead by up to 72.9%, achieving a 3.69 xtraining speedup while improving fine-tuning accuracy by 3.50%.

Visual-language grounding aims to establish semantic correspondences between natural language and visual entities, enabling models to accurately identify and localize target objects based on textual instructions. Existing VLG approaches focus on coarse-grained, object-level localization, while traditional robotic grasping methods rely predominantly on geometric cues and lack language guidance, which limits their applicability in language-driven manipulation scenarios. To address these limitations, we propose the RealVLG framework, which integrates the RealVLG-11B dataset and the RealVLG-R1 model to unify real-world visual-language grounding and grasping tasks. RealVLG-11B dataset provides multi-granularity annotations including bounding boxes, segmentation masks, grasp poses, contact points, and human-verified fine-grained language descriptions, covering approximately 165,000 images, over 800 object instances, 1.3 million segmentation, detection, and language annotations, and roughly 11 billion grasping examples. Building on this dataset, RealVLG-R1 employs Reinforcement Fine-tuning on pretrained large-scale vision-language models to predict bounding boxes, segmentation masks, grasp poses, and contact points in a unified manner given natural language instructions. Experimental results demonstrate that RealVLG supports zero-shot perception and manipulation in real-world unseen environments, establishing a unified semantic-visual multimodal benchmark that provides a comprehensive data and evaluation platform for language-driven robotic perception and grasping policy learning. All data and code are publicly available at https://github.com/lif314/RealVLG-R1.

In safety-critical industrial applications, accurately identifying causal relationships and quantifying uncertainty is essential for tasks such as root cause analysis, feature selection, and process optimization. Traditional causal discovery methods inadequately handle nonlinearities and complex uncertainties prevalent in industrial sensor data. To address this, we introduce a novel causal discovery framework that integrates Polynomial Chaos Expansion (PCE) representations of stochastic noise into structural equations. This method effectively captures complex nonlinear couplings and arbitrary noise distributions characteristic of industrial data. We rigorously prove the identifiability of causal structures under mild sparsity conditions on the chaos coefficients, extending linear non-Gaussian acyclic model (LiNGAM) identifiability results. Extensive experiments demonstrate superior accuracy, robustness under non-Gaussian noise conditions, and practical uncertainty quantification. This framework presents a principled, interpretable, and computationally feasible approach to causal analysis in nonlinear uncertain industrial environments.

We propose POLAR, a novel radar-guided depth estimation method that introduces polynomial fitting to efficiently transform scaleless depth predictions from pretrained monocular depth estimation (MDE) models into metric depth maps. Unlike existing approaches that rely on complex architectures or expensive sensors, our method is grounded in a fundamental insight: although MDE models often infer reasonable local depth structure within each object or local region, they may misalign these regions relative to one another, making a linear scale and shift (affine) transformation insufficient given three or more of these regions. To address this limitation, we use polynomial coefficients predicted from cheap, ubiquitous radar data to adaptively adjust predictions non-uniformly across depth ranges. In this way, POLAR generalizes beyond affine transformations and is able to correct such misalignments by introducing inflection points. Importantly, our polynomial fitting framework preserves structural consistency through a novel training objective that enforces local monotonicity via first-derivative regularization. POLAR achieves state-of-the-art performance across three datasets, outperforming existing methods by an average of 24.9% in MAE and 33.2% in RMSE, while also achieving state-of-the-art efficiency in terms of latency and computational cost.

In this paper, we introduce Geometric Algebra-Informed 3D Gaussian Splatting (GAI-GS), a framework for wireless modeling that couples 3D Gaussian splatting with a geometric algebra-based attention mechanism to explicitly model ray-object interactions in complex propagation environments. GAI-GS encodes joint spatial-electromagnetic (EM) relations into token representations, enabling scene-level aggregation within a unified, end-to-end neural architecture. This design grounds wireless ray propagation in electromagnetic principles, allowing token interactions to model key effects such as multipath, attenuation, and reflection/diffraction. Through extensive evaluations on multiple real-world indoor datasets, GAI-GS consistently surpasses current baselines across various wireless tasks.