CVPR 2026 Papers

Simulation of cellular morphology change has long been a fundamental task in quantitative biology and high-throughput screening, with the potential to accelerate therapeutic development and elucidate disease mechanisms beyond empirical clinical practice. However, the vast perturbation space poses challenges to the discriminative formulation, and existing generative approaches tend to concentrate on the same trajectory subspace, making their generated paths prone to drift. In this paper, we propose a novel Steered Diffusion Bridge approach, named SimuSDB, to define deterministic probabilistic trajectories between two distinct state domains for cell response generation. We first extend the diffusion bridge paradigm to maintain stochasticity and diversity in interpolation trajectories by introducing Brownian bridges. Then, SimuSDB generates cell morphologies that comply with phenotypic constraints, while allowing the latter to explicitly guide the generative process. For the inference stage, we formalize the rule-guided sample generation task as an optimal control problem within a stochastic dynamical system. This way, the generative model can achieve analytically tractable optimal control strategies and steered generation without collapsing toward the trajectory of the same data subspace. Comprehensive experiments demonstrate the superior performance of SimuSDB across various applications, including chemical perturbation and genetic perturbation.

Recent advances in zero-shot learning (ZSL) have demonstrated the potential of generative models. Typically, generative ZSL synthesizes visual features conditioned on semantic prototypes to model the data distribution of unseen classes, followed by training a classifier on the synthesized data. However, the synthesized features often remain task-agnostic, leading to degraded performance. Moreover, inferring a faithful distribution from semantic prototypes alone is insufficient for classes that are semantically similar but visually distinct. To address these and advance ZSL, we propose RLVC, an outcome-reward reinforcement learning RL framework with visual cues for generative ZSL. At its core, RL empowers the generative model to self-evolve, implicitly enhancing its generation capability. In particular, RLVC updates the generative model using an outcome-based reward, encouraging the synthesis of task-relevant features. Furthermore, we introduce class-wise visual cues that (i) align synthesized features with visual prototypes and (ii) stabilize the RL training updates. For the training process, we present a novel cold-start strategy. Comprehensive experiments and analyses on three prevalent ZSL benchmarks demonstrate that RLVC achieves state-of-the-art results with a 4.7% gain.

The Mean Flow Matching algorithm is the state-of-the-art for one-step generative models. Building on this idea, we propose the Stable Mean Flow algorithm and introduce a Lyapunov-inspired stability regularizer that enforces local non-expansivity of the single-step transport map. This design guarantees uniqueness of characteristics and bounds trajectory drift. We conduct experiments that show improved output quality and convergence speed over Mean Flow. Moreover, we establish explicit upper bounds on error growth for both one-step and multi-step generation.

We present ARTrack-AC, a new step in the autoregressive tracking paradigm that introduces adaptive capacity inference to achieve both temporal consistency and dynamic efficiency. While existing autoregressive trackers predict object states sequentially with fixed inference capacity, they fail to accommodate the fluctuating temporal difficulty of real videos. ARTrack-AC addresses this limitation by equipping the tracker with the ability to modulate its inference capacity over time. A diffusion-based difficulty estimator anticipates the stability of upcoming segments, guiding a controller to switch between an accurate (high-capacity) and an efficient (low-capacity) mode while maintaining autoregressive consistency. This system-level autoregression extends conventional sequence modeling beyond "what to predict" toward "how to predict," forming a self-regulated tracking process that aligns inference cost with temporal complexity. Despite its simplicity, ARTrack-AC achieves state-of-the-art accuracy-speed trade-off on major benchmarks--66.7% AUC on LaSOT and 47.5% AUC on LaSOText--running 2.9xfaster than its predecessor.

Our goal is to train a generative model of 3D hand motions, conditioned on natural language descriptions specifying motion characteristics such as handshapes, locations, finger/hand/arm movements. To this end, we automatically build pairs of 3D hand motions and their associated textual labels with unprecedented scale. Specifically, we leverage a large-scale sign language video dataset, along with noisy pseudo-annotated sign categories, which we translate into hand motion descriptions via an LLM that utilizes a dictionary of sign attributes, as well as our complementary motion-script cues. This data enables training a text-conditioned hand motion diffusion model (HandMDM), that is robust across domains such as unseen sign categories from the same sign language, but also signs from another sign language and non-sign hand movements. We contribute extensive experimental investigation of these scenarios and will make our trained models and data publicly available to support future research in this relatively new field.

Parameter Recombination (PR) methods aim to efficiently compose the weights of a neural network, and encompasses tasks like Parameter-Efficient FineTuning (PEFT) and Model Compression (MC), among others. Most methods typically focus on one application of PR, which can make composing them challenging. For example, when deploying a large model you may wish to compress the model and also quickly adapt to new settings. However, PEFT methods often can still contain millions of parameters. This may be small compared to the original model size, but can be problematic in resource constrained deployments like edge devices, where they take a larger portion of the compressed model's parameters. To address this, we present Coefficient-gated weight Recombination by Interpolated Shared basis Projections (\method ), a general approach that can address multiple PR tasks within the same framework, which can enable seamless integration. It accomplishes this by using a factorization process that decomposes pretrained weights into basis matrices and their component projections. Sharing these basis matrices across layers and adjusting its size enables us to perform MC, whereas the small size of the projection weights (fewer than 200 in some experiments) enables \method support PEFT. Experiments on ViT models show \method outperforms methods from prior work capable of dual-task applications by 4-5% while also outperforming the state-of-the-art in PEFT by 1.5% and PEFT+MC combinations by almost 1%.

Virtual try-on (VTON) has advanced single-garment visualization, yet real-world fashion centers on full outfits with multiple garments, accessories, fine-grained categories, layering, and diverse styling, remaining beyond current VTON systems. Existing datasets are category-limited and lack outfit diversity. We introduce Garments2Look, the first large-scale multimodal dataset for outfit-level VTON, comprising 80K many-garments-to-one-look pairs across 40 major categories and 300+ fine-grained subcategories. Each pair includes an outfit with 3-12 reference garment images (Average 4.48), a model image wearing the outfit, and detailed item and try-on textual annotations. To balance authenticity and diversity, we propose a synthesis pipeline. It involves heuristically constructing outfit lists before generating try-on results, with the entire process subjected to strict automated filtering and human validation to ensure data quality. To probe task difficulty, we adapt SOTA VTON methods and general-purpose image editing models to establish baselines. Results show current methods struggle to try on complete outfits seamlessly and to infer correct layering and styling, leading to misalignment and artifacts.

Many classic opera videos exhibit poor visual quality due to the limitations of early filming equipment and long-term degradation during storage. Although real-world video super-resolution (RWVSR) has achieved significant advances in recent years, directly applying existing methods to degraded opera videos remains challenging. The difficulties are twofold. First, accurately modeling real-world degradations is complex: simplistic combinations of classical degradation kernels fail to capture the authentic noise distribution, while methods that extract real noise patches from external datasets are prone to style mismatches that introduce visual artifacts. Second, current RWVSR methods, which rely solely on degraded image features, struggle to reconstruct realistic and detailed textures due to a lack of high-level semantic guidance. To address these issues, we propose a Text-guided Dual-Branch Opera Video Super-Resolution (TextOVSR) network, which introduces two types of textual prompts to guide the super-resolution process. Specifically, degradation-descriptive text, derived from the degradation process, is incorporated into the negative branch to constrain the solution space. Simultaneously, content-descriptive text is incorporated into a positive branch and our proposed Text-Enhanced Discriminator (TED) to provide semantic guidance for enhanced texture reconstruction. Furthermore, we design a Degradation-Robust Feature Fusion (DRF) module to facilitate cross-modal feature fusion while suppressing degradation interference. Experiments on our OperaLQ benchmark show that TextOVSR outperforms state-of-the-art methods both qualitatively and quantitatively. The code is available at https://github.com/ChangHua0/TextOVSR.

Vision-Language Models (VLMs) typically assume a uniform spatial fidelity across the entire field of view of visual inputs, dedicating equal precision to even the uninformative regions. By contrast, human vision is neither uniform nor static. It is adaptive, selective, and resource-efficient. In light of this, we present the first systematic analysis of bio-inspired visual representation methods, providing insights for more efficient and adaptive VLMs. We propose LLMind (Looking Like the Mind), a novel training-free framework that mimics foveated encoding and cortical magnification in human vision to achieve adaptive, efficient representations for VLMs under tight pixel budgets. Our key idea is to explore a Bio-inspired Adaptive Sampling Strategy (BASS). This empowers us to design a Mobius-parameterized module that performs non-uniform sampling while preserving global scene structure. On top of BASS, we introduce closed-loop semantic feedback (CSF) via test-time adaptation to align the perceptual saliency with the textual information from the frozen VLM. We evaluate LLMind against uniform and other sampling baselines across diverse scene-level and region-guided visual question answering (VQA) benchmarks. The results show that ours achieves dramatic gains, with average improvements by +20% on VQAv2, +38% on Seed-Bench, and +37% on A-OKVQA compared to uniform sampling under tight pixel budgets. More surprisingly, results reveal that LLMind can retain up to 82%, 92% and 97% of the full-resolution performance with only 1%, 3% and 5% of the pixels, respectively. Moreover, LLMind is lightweight, plug-and-play, and compatible with existing VLMs without requiring architectural changes.

We present ReMoT, a unified training paradigm to systematically address the fundamental shortcomings of VLMs in spatio-temporal consistency--a critical failure point in navigation, robotics, and autonomous driving. ReMoT integrates two core components: (i) A rule-based automatic framework that generates ReMoT-16K, a large-scale (16.5K triplets) motion-contrast dataset derived from video meta-annotations, surpassing costly manual or model-based generation. (ii) Group Relative Policy Optimization, which we empirically validate, yields optimal performance and data efficiency for learning this contrastive reasoning, far exceeding standard Supervised Fine-Tuning. We also construct the first benchmark for fine-grained motion contrast triplets to measure a VLM's discrimination of subtle motion attributes (e.g., opposing directions). The resulting model achieves SOTA performance on our new benchmark and multiple standard VLM benchmarks, culminating in a remarkable 25.1 performance leap on spatio-temporal reasoning tasks.

Blind-spot networks (BSNs) enable self-supervised image denoising by preventing access to the target pixel, allowing clean signal estimation without ground-truth supervision. However, this approach assumes pixel-wise noise independence, which is violated in real-world sRGB images due to spatially correlated noise from the camera's image signal processing (ISP) pipeline. While several methods employ downsampling to decorrelate noise, they alter noise statistics and limit the network's ability to utilize full contextual information. In this paper, we propose the Triangular-Masked Blind-Spot Network (TM-BSN), a novel blind-spot architecture that accurately models the spatial correlation of real sRGB noise. This correlation originates from demosaicing, where each pixel is reconstructed from neighboring samples with spatially decaying weights, resulting in a diamond-shaped pattern. To align the receptive field with this geometry, we introduce a triangular-masked convolution that restricts the kernel to its upper-triangular region, creating a diamond-shaped blind spot at the original resolution. This design excludes correlated pixels while fully leveraging uncorrelated context, eliminating the need for downsampling or post-processing. Furthermore, we use knowledge distillation to transfer complementary knowledge from multiple blind-spot predictions into a lightweight U-Net, improving both accuracy and efficiency. Extensive experiments on real-world benchmarks demonstrate that our method achieves state-of-the-art performance, significantly outperforming existing self-supervised approaches. Our code is available at https://github.com/parkjun210/TM-BSN.

Autoregressive mesh generation has gained attention by tokenizing meshes into sequences and training models in a language-modeling fashion. However, existing approaches suffer from two fundamental limitations: (i) low tokenization efficiency, which yields long token sequences and prevents scaling to high-poly meshes, and (ii) absence of geometry-aware guidance, as generation is conditioned only on global shape embeddings rather than local surface cues. We introduce MeshWeaver, an autoregressive framework that treats mesh generation as a surface weaving process by directly predicting the next vertex instead of independent coordinates. At its core is a multi-level sparse-voxel encoder that injects geometric context into the generative process in three complementary ways: providing voxel features as vertex representations, guiding token prediction via cross-attention to voxel features, and serving as a structural scaffold that constrains generation around the input surface. Our hierarchical design enables coarse-to-fine vertex prediction in a single decoding step, while tightly coupling the generative model with 3D geometry. Extensive experiments demonstrate that MeshWeaver achieves a state-of-the-art compression ratio of 18%, can generate meshes with up to 16K faces, and significantly improves geometric fidelity over prior approaches.

Current multimodal image retrieval benchmarks focus on relatively simple queries where target images are either described directly or by simple composition with an input image. When retrieval requires complex reasoning to determine the target image, the task becomes significantly more challenging, yet standardized benchmarks for this setting do not exist. To fill this gap, we introduce RMIR, a benchmark dataset of 1634 queries requiring reasoning across three categories: functional (object affordances), temporal (time-based relationships), and causal (cause-effect reasoning). Each query combines visual and textual inputs that demand robust visual understanding together with logical inference, beyond surface-level matching, to identify correct target images. In addition to the dataset itself, we present a pipeline to generate the dataset, which can be used to generate additional reasoning-intensive retrieval data at scale. Evaluation of state-of-the-art models on RMIR reveals significant performance gaps, with the best model achieving only 46.53% recall@20 averaged across reasoning categories. Our evaluation also shows that generative embedding models with explicit reasoning substantially outperform discriminative approaches, with reasoning-aware training proving more impactful than model scale. Our systematic analysis exposes fundamental limitations in current multimodal retrieval systems and establishes RMIR as a challenging testbed for developing multimodal, reasoning-capable retrieval models. Our dataset and code are available at https://github.com/amazon-science/rmir

Face recognition (FR) technologies are increasingly used to power large-scale image retrieval systems, raising serious privacy concerns. Services like Clearview AI and PimEyes allow anyone to upload a facial photo and retrieve a large amount of online content associated with that person. This not only enables identity inference but also exposes their digital footprint, such as social media activity, private photos, and news reports, often without their consent. In response to this emerging threat, we propose Protego, a user-centric privacy protection method that safeguards facial images from such retrieval-based privacy intrusions. Protego encapsulates a user's 3D facial signatures into a pose-invariant 2D representation, which is dynamically deformed into a natural-looking 3D mask tailored to the pose and expression of any facial image of the user, and applied prior to online sharing. Motivated by a critical limitation of existing methods, Protego amplifies the sensitivity of FR models so that protected images cannot be matched even among themselves. Experiments show that Protego significantly reduces retrieval accuracy across a wide range of black-box FR models and performs at least 2x better than existing methods. It also offers unprecedented visual coherence, particularly in video settings where consistency and natural appearance are essential. Overall, Protego contributes to the fight against the misuse of FR for mass surveillance and identity tracing.

Editable high-fidelity 4D scenes are crucial for autonomous driving, as they can be applied to end-to-end training and closed-loop simulation. However, existing reconstruction methods are primarily limited to replicating observed scenes and lack the capability for diverse weather simulation. While image-level weather editing methods tend to introduce scene artifacts and offer poor controllability over the weather effects. To address these limitations, we propose WeatherCity, a novel framework for 4D urban scene reconstruction and weather editing. Specifically, we leverage a text-guided image editing model to achieve flexible editing of image weather backgrounds. To tackle the challenge of multi-weather modeling, we introduce a novel weather Gaussian representation based on shared scene features and dedicated weather-specific decoders. This representation is further enhanced with a content consistency optimization, ensuring coherent modeling across different weather conditions. Additionally, we design a physics-driven model that simulates dynamic weather effects through particles and motion patterns. Extensive experiments on multiple datasets and various scenes demonstrate that WeatherCity achieves flexible controllability, high fidelity, and temporal consistency in 4D reconstruction and weather editing. Our framework not only enables fine-grained control over weather conditions (e.g., light rain and heavy snow) but also supports object-level manipulation within the scene. Codes are released at https://github.com/IRMVLab/WeatherCity.

Existing autonomous driving systems rely on onboard sensors (cameras, LiDAR, IMU, etc) for environmental perception. However, this paradigm is limited by the drive-time perception horizon and often fails under limited view scope, occlusion or extreme conditions such as darkness and rain. In contrast, human drivers are able to recall road structure even under poor visibility. To endow models with this recall ability, we propose the spatial retrieval paradigm, introducing offline retrieved geographic images as an additional input. These images are easy to obtain from offline caches (e.g., Google Maps or stored autonomous driving datasets) without requiring additional sensors, making it a plug-and-play extension for existing AD tasks. For experiments, we first extend the nuScenes dataset with geographic images retrieved via Google Maps APIs and align the new data with ego-vehicle trajectories. We establish baselines across five core autonomous driving tasks: object detection, online mapping, occupancy prediction, end-to-end planning, and generative world modeling. Extensive experiments show that the extended modality could enhance the performance of certain tasks. We will open-source dataset curation code, data, and benchmarks for further study of this new autonomous driving paradigm.

Chain-of-Thought (CoT) has recently shown encouraging progress in the vision language model. However, the pure-vision CoT (i.e., chain-of-vision) has been underexplored in visual in-context learning. In this paper, we introduce Diffusion Guided Chain-of-Vision, which integrates an explicit chain-of-thought process into autoregressive vision models through vision prior from pre-trained diffusion models. Concretely, we find that pre-trained diffusion models induce a reliable probability flow in image space, where intermediate images sampled along this flow exhibit visual coherence and serve as task-free, chain-of-vision supervision for pure-vision autoregressive models. Extensive experiments on diverse vision tasks and multi-scale models validate the effectiveness of our proposed method for visual in-context learning. Code and dataset will be publicly available.

3D super-resolution (3DSR) aims to reconstruct high-resolution (HR) 3D scenes from low-resolution (LR) multi-view images. Existing methods rely on dense LR inputs and per-scene optimization, which restricts the high-frequency priors for constructing HR 3D Gaussian Splatting (3DGS) to those inherited from pretrained 2D super-resolution (2DSR) models. This severely limits reconstruction fidelity, cross-scene generalization, and real-time usability. We propose to reformulate 3DSR as a direct feed-forward mapping from sparse LR views to HR 3DGS representations, enabling the model to autonomously learn 3D-specific high-frequency geometry and appearance from large-scale, multi-scene data. This fundamentally changes how 3DSR acquires high-frequency knowledge and enables robust generalization to unseen scenes. Specifically, we introduce SR3R, a feed-forward framework that directly predicts HR 3DGS representations from sparse LR views via the learned mapping network. To further enhance reconstruction fidelity, we introduce Gaussian offset learning and feature refinement, which stabilize reconstruction and sharpen high-frequency details. SR3R is plug-and-play and can be paired with any feed-forward 3DGS reconstruction backbone: the backbone provides an LR 3DGS scaffold, and SR3R upscales it to an HR 3DGS. Extensive experiments across three 3D benchmarks demonstrate that SR3R surpasses state-of-the-art (SOTA) 3DSR methods and achieves strong zero-shot generalization, even outperforming SOTA per-scene optimization methods on unseen scenes.

The application of Large Multimodal Models (LMMs) to long-form video understanding is constrained by limited context lengths and the computationally prohibitive cost of processing dense video tokens. Consequently, recent research has focused on query-aware frame selection, methods that often incur significant computational overhead. This paper challenges the assumption that such complex search mechanisms are universally necessary. We first identify and validate a query typology distinguishing between global query and localized query. We demonstrate that while uniform sampling is both effective and efficient for global queries, localized queries indeed necessitate query-aware selection for optimal performance. Building on this insight, we propose DIG, a training-free frame selection framework that adapts its strategy based on the query type. Specifically, DIG employs efficient uniform sampling for global queries while activating a specialized pipeline to extract query-relevant frames for localized queries. Experiments on three long-form video understanding benchmarks demonstrate that DIG consistently outperforms existing baselines and robustly improves LMM performance, even when scaling the input frame count to 256.