CVPR 2026 论文列表

Existing diffusion codecs typically build on text-to-image diffusion foundation models like Stable Diffusion.However, text conditioning is suboptimal from a compression perspective, hindering the potential of downstream diffusion codecs, particularly at ultra-low bitrates.To address it, we introduce CoD, the first Compression-oriented Diffusion foundation model, trained from scratch to enable end-to-end optimization of both compression and generation. CoD is not a fixed codec but a general foundation model designed for various diffusion-based codecs.It offers several advantages: High compression efficiency, replacing Stable Diffusion with CoD in downstream codecs like DiffC achieves SOTA results, especially at ultra-low bitrates (e.g., 0.0039 bpp); Low-cost and reproducible training, 300x faster training than Stable Diffusion ( 20 vs. 6,250 A100 GPU days) on entirely open image-only datasets; Providing new insights, e.g., We find pixel-space diffusion can achieve VTM-level PSNR with high perceptual quality and can outperform GAN-based codecs using fewer parameters.We hope CoD lays the foundation for future diffusion codec research. Codes are released at https://github.com/microsoft/GenCodec/tree/main/CoD.

Training-free open-vocabulary semantic segmentation (OVSS) promises rapid adaptation to new label sets without retraining. Yet, many methods rely on heavy post-processing or handle text and vision in isolation, leaving cross-modal geometry underutilized. Others introduce auxiliary vision backbones or multi-model pipelines, which increase complexity and latency while compromising design simplicity.We present PEARL, \underline P rocrust\underline e s \underline a lignment with text-awa\underline r e \underline L aplacian propagation, a compact two-step inference that follows an align-then-propagate principle. The Procrustes alignment step performs an orthogonal projection inside the last self-attention block, rotating keys toward the query subspace via a stable polar iteration. The text-aware Laplacian propagation then refines per-pixel logits on a small grid through a confidence-weighted, text-guided graph solve: text provides both a data-trust signal and neighbor gating, while image gradients preserve boundaries. In this work, our method is fully training-free, plug-and-play, and uses only fixed constants, adding minimal latency with a small per-head projection and a few conjugate-gradient steps. Our approach, PEARL, sets a new state-of-the-art in training-free OVSS without extra data or auxiliary backbones across standard benchmarks, achieving superior performance under both with-background and without-background protocols.

Although convolutional neural networks (CNNs) have continued to evolve in recent years, Transformers have become increasingly popular in the field of computer vision. In this work, we introduce a mechanism for CNNs to aggregate information based on content similarity--an ability analogous to the self-attention mechanism. We achieve this by proposing a sorting operation that transforms the feature similarity between tokens into relative positional information: specifically, rearranging tokens such that tokens with higher feature similarity are placed closer together. This approach allows convolution operations to be indirectly transformed into an aggregation mode driven by content similarity. Experiments show that our proposed model, named Ego, achieves promising results across various tasks, underscoring the untapped potential of CNNs. Code and models are publicly available at https://github.com/essenceoftheworld/ego.

Conventional few-shot medical image segmentation (FSMIS) approaches face performance bottlenecks that hinder broader clinical applicability. Although the Segment Anything Model (SAM) exhibits strong category-agnostic segmentation capabilities, its direct application to medical images often leads to over-segmentation due to ambiguous anatomical boundaries. In this paper, we reformulate SAM-based FSMIS as a prompt localization task and propose FoB (Focus on Background), a background-centric prompt generator that provides accurate background prompts to constrain SAM's over-segmentation. Specifically, FoB bridges the gap between segmentation and prompt localization by category-agnostic generation of support background prompts and localizing them directly in the query image. To address the challenge of prompt localization for novel categories, FoB models rich contextual information to capture foreground-background spatial dependencies. Moreover, inspired by the inherent structural patterns of background prompts in medical images, FoB models this structure as a constraint to progressively refine background prompt predictions. Experiments on three diverse medical image datasets demonstrate that FoB outperforms other baselines by large margins, achieving state-of-the-art performance on FSMIS, and exhibiting strong cross-domain generalization.

As a challenge vision-language task, Composed Image Retrieval (CIR) aims to integrate information from a bi-modal query (image + text) to retrieve target images. While supervised CIR has achieved notable success in domain-specific scenarios, its reliance on manually annotated triplets restricts its scalability and application. Zero-shot CIR alleviates this by leveraging unlabeled data or automatically collected triplets, yet it often suffers from an intractable domain gap. To this end, we shift the focus to developing robust CIR models under limited labeled data and propose Domain-Adaptive In-context Generation (DAIG), which adapts the in-context capability of a pretrained Text-to-Image (T2I) model to the target domain and the CIR task using few-shot samples and then transforms the LLM-generated textual triplets into unbiased CIR triplets as additional training data. After that, we present a two-stage framework applicable to any supervised CIR approach. The first stage, Distributionally Robust Synthetic Pretraining (DRSP), perturbs visual features to expand the distribution of synthetic data and improve training robustness on it. The second stage, Fine-grained Real-world Adaptation (FRA), fine-tunes on manually annotated triplets by imposing an angular margin on matching pairs to facilitate fine-grained learning. Experiments on two benchmarks validate the effectiveness of our method, i.e., under both few-shot and fully supervised CIR settings, DAIG yields substantial performance gains over CLIP4CIR, BLIP4CIR, and SPRC. The code is available at \href https://github.com/JThuge/DAIG \color blue!37https://github.com/JThuge/DAIG .

This paper presents FluxMem, a training-free framework for efficient streaming video understanding. FluxMem adaptively compresses redundant visual memory through a hierarchical, two-stage design: (1) a Temporal Adjacency Selection (TAS) module removes redundant visual tokens across adjacent frames, and (2) a Spatial Domain Consolidation (SDC) module further merges spatially repetitive regions within each frame into compact representations. To adapt effectively to dynamic scenes, we introduce a self-adaptive token compression mechanism in both TAS and SDC, which automatically determines the compression rate based on intrinsic scene statistics rather than manual tuning. Extensive experiments demonstrate that FluxMem achieves new state-of-the-art results on existing online video benchmarks, reaching 76.4 on StreamingBench and 67.2 on OVO-Bench under real-time settings, while reducing latency by 69.9% and peak GPU memory by 34.5% on OVO-Bench. Furthermore, it maintains strong offline performance, achieving 73.1 on MLVU while using 65% fewer visual tokens.

Detecting small aerial targets (SATs) in long-range LiDAR is challenging because motion causes extreme variations in point density, breaking fixed-voxel and static-threshold assumptions in standard 3D detection and tracking. To address the challenges, we introduce A3PRL, an RL-driven adaptive perception framework that closes the loop between LiDAR sensing and tracking. A3PRL uses a sparsity-aware proposal stage with Temporal Dispersion Signatures and velocity-change cues, and a lightweight 5D policy that jointly adjusts voxel resolution, detection sensitivity, and association gating from label-free statistics of sparsity, foreground acceptance, and track stability. The policy is trained with privileged ground-truth trajectories to optimize a reward balancing geometric accuracy, temporal stability, and regularized acceptance, but runs fully label-free at test time. On the public MMAUD benchmark, training on V1 and evaluating on unseen V2/V3, A3PRL reduces 3D localization error by about 19% over its non-RL counterpart and consistently outperforms LiDAR-only and multimodal baselines under both day and night conditions. The same policy is able to transfer to other SAT datasets with heterogeneous scan patterns, maintaining accurate and stable trajectories.

Instruction-based video editing aims to modify an input video according to a natural-language instruction while preserving content fidelity and temporal coherence. However, existing diffusion-based approaches are often trained on paired data of simple editing operations, which fundamentally limits their ability to generalize to diverse and complex, real-world instructions. To address this generalization gap, we propose VIVA, a scalable framework for instruction-based video editing that leverages VLM-guided encoding and reward optimization. First, we introduce a VLM-based instructor that encodes the textual instruction, the first frame of the source video, and an optional reference image into visually-grounded instruction representations, providing fine-grained spatial and semantic context for the diffusion transformer backbone. Second, we propose a post-training stage, Edit-GRPO, which adapts Group Relative Policy Optimization to the domain of video editing, directly optimizing the model for instruction-faithful, content-preserving, and aesthetically pleasing edits using relative rewards. Furthermore, we propose a data construction pipeline designed to synthetically generate diverse, high-fidelity paired video-instruction data of basic editing operations. Extensive experiments show that VIVA achieves superior instruction following, generalization, and editing quality over state-of-the-art methods.

Although EDM aims to unify the design space of diffusion models, its reliance on fixed Gaussian noise prevents it from explaining emerging flow-based methods that diffuse arbitrary noise. Moreover, our study reveals that EDM's forcible injection of Gaussian noise has adverse effects on image restoration task, as it corrupts the degraded images, overextends the restoration distance, and increases the task's complexity. To interpret diverse methods for handling distinct noise patterns within a unified theoretical framework and to minimize the restoration distance, we propose EDA, which Elucidates the Design space of Arbitrary-noise diffusion models. Theoretically, EDA expands noise pattern flexibility while preserving EDM's modularity, with rigorous proof that increased noise complexity introduces no additional computational overhead during restoration. EDA is validated on three representative medical image denoising and natural image restoration tasks: MRI bias field correction (global smooth noise), CT metal artifact removal (global sharp noise) and natural image shadow removal (local boundary-aware noise). With only 5 sampling steps, competitive results against specialized methods across medical and natural tasks demonstrate EDA's strong generalization capability for image restoration. Code is available at: https://github.com/PerceptionComputingLab/EDA.

Chart Question Answering (CQA) benchmarks are critical for evaluating Multimodal Large Language Models (MLLMs) on visual data reasoning. Existing benchmarks focus mainly on final-answer correctness, ignoring intermediate reasoning steps and the propagation of errors in multi-step processes. To address this, we introduce ChartR, a benchmark designed to assess both the accuracy and robustness of reasoning in chart-understanding tasks. Each question is decomposed into 4-10 sub-questions covering key reasoning types, and each chart includes four visually perturbed variants (blurred, noise-added, watermark-added, annotation-removed) to systematically evaluate robustness. ChartR contains 200 base charts, 800 variants, 1,652 questions, and 8,260 image-question pairs. We further propose a comprehensive evaluation framework with eight metrics that evaluate reasoning-chain accuracy, robustness under visual perturbations, and enable analysis of potential error propagation patterns. Experiments on twelve MLLMs, including general-purpose and chart-specialized models, reveal low reasoning reliability, early-step errors that may propagate, value extraction as the primary bottleneck, and sharp performance drops under perturbations, highlighting reliance on textual cues over true visual understanding.

We present CUBE (Control-based Unified B-Splinie Encoding), a new geometric representation for human faces that combines B-spline volumes with learned features, and demonstrate its use as a decoder for 3D scan registration and monocular 3D face reconstruction. Unlike existing B-spline representations with 3D control points, CUBE is parametrized by a lattice (e.g., 8 x 8 x 8) of high-dimensional control features, increasing the model's expressivity. These features define a continuous, two-stage mapping from a 3D parametric domain to 3D Euclidean space via an intermediate feature space. First, high-dimensional control features are locally blended using the B-spline bases, yielding a high-dimensional feature vector whose first three values define a 3D base mesh. A small MLP then processes this feature vector to predict a residual displacement from the base shape, yielding the final refined 3D coordinates. To reconstruct 3D surfaces in dense semantic correspondence, CUBE is queried at 3D coordinates sampled from a fixed template mesh. Crucially, CUBE retains the local support property of traditional B-spline representations, enabling local surface editing by updating individual control features. We demonstrate the strengths of this representation by training transformer-based encoders to predict CUBE's control features from unstructured point clouds and monocular images, achieving state-of-the-art scan registration results compared to recent baselines.

Constructing a 4D language field that supports open-vocabulary queries is essential for semantic perception and interaction in dynamic environments. Existing 4D Gaussian-based approaches face two major challenges. First, the assumption of a static identity per Gaussian leads to semantic inconsistency, as motion fields warp Gaussians across object boundaries over time, causing oscillating identity assignments. Second, current methods typically model dynamic semantics as a set of discrete, predefined state prototypes, which fail to capture dynamic continuity and delineate fine-grained temporal boundaries. To address these issues, we propose LangField4D, a novel 4D Gaussian framework that jointly models spatio-temporal identity and semantics in a unified and continuous representation. We introduce an Identity-Adaptive Gaussian Grouping module that assigns each Gaussian a learnable adaptation feature to dynamically capture its object affiliation, ensuring consistent semantic tracking across time. Building upon this affiliation structure, we further design a Continuous Spatio-Temporal Semantic Learning mechanism based on a Tetraplane representation, which encodes both time-invariant and time-varying semantics within a continuous latent space. Extensive experiments on dynamic scene benchmarks demonstrate that we achieve state-of-the-art performance on both time-agnostic and time-sensitive open-vocabulary query tasks.

Visual concept personalization aims to transfer only specific image attributes, such as identity, expression, lighting, and style, into unseen contexts. However, existing methods rely on holistic embeddings from general-purpose image encoders, which entangle multiple visual factors and make it difficult to isolate a single attribute. This often leads to information leakage and incoherent synthesis. To address this limitation, we introduce Omni-Attribute, the first open-vocabulary image attribute encoder designed to learn high-fidelity, attribute-specific representations. Our approach jointly designs the data and model: (i) we curate semantically linked image pairs annotated with positive and negative attributes to explicitly teach the encoder what to preserve or suppress; and (ii) we adopt a dual-objective training paradigm that balances generative fidelity with contrastive disentanglement. The resulting embeddings prove effective for open-vocabulary attribute retrieval, personalization, and compositional generation, achieving state-of-the-art performance across multiple benchmarks.

Change detection (CD) is a fundamental task for monitoring and analysing land cover dynamics. While recent high performance models and high quality datasets have significantly advanced the field, a critical limitation persists. Current models typically acquire limited knowledge from single-type annotated data and cannot concurrently leverage diverse binary change detection (BCD) and semantic change detection (SCD) datasets. This constraint leads to poor generalisation and limited versatility. The recent advancements in Multimodal Large Language Models (MLLMs) introduce new possibilities for a unified CD framework. We leverage the language priors and unification capabilities of MLLMs to develop UniChange, the first MLLM-based unified change detection model. UniChange integrates generative language abilities with specialised CD functionalities. We introduce three special tokens: [T1], [T2], and [CHANGE], utilising their embeddings as the key to query variations. This approach successfully accommodates both BCD and SCD tasks. Furthermore, UniChange utilises text prompts to guide the identification of change categories, eliminating the reliance on predefined classification heads. This design allows UniChange to effectively acquire knowledge from multi-source datasets, even when their class definitions conflict. Experiments on four public benchmarks (WHU-CD, S2Looking, LEVIR-CD+, and SECOND) demonstrate SOTA performance, achieving IoU scores of 90.41, 53.04, 78.87, and 57.62, respectively, surpassing all previous methods. The code is available at https://github.com/NKU-HLT/UniChange.

Stereo foundation models achieve strong zero-shotgeneralization but remain computationally prohibitive forreal-time applications. Efficient stereo architectures, on the other hand, sacrificerobustness for speed and require costly per-domain fine-tuning.To bridge this gap, we present Fast-FoundationStereo, a family of architectures that achieve, for the first time, strong zero-shot generalization at real-time frame rate. We employ a divide-and-conquer acceleration strategy with three components: (1) knowledge distillation to compress the hybrid backbone into a single efficient student; (2) blockwise neural architecture search for automatically discovering optimal cost filtering designs under latency budgets, reducing search complexity exponentially; and (3) structured pruning for eliminating redundancy in the iterative refinement module. Furthermore, we introduce an automatic pseudo-labeling pipeline used to curate 1.4M in-the-wild stereo pairs to supplement synthetic training data and facilitate knowledge distillation. The resulting model can run over 10x faster than FoundationStereo while closely matching its zero-shot accuracy, thus establishing a new state-of-the-art among real-time methods. Project page: https://nvlabs.github.io/Fast-FoundationStereo

Appropriate parameter initialization is crucial for reducing the training cost of deep neural networks. Graph HyperNetworks (GHN) have emerged as a promising approach for initializing diverse architectures, with recent methods such as Task-Aware Learngene (TAL) further attempting to leverage pre-trained model knowledge via soft label supervision. However, such indirect supervision fails to fully exploit the rich information encoded in pre-trained weights. We propose **P**arameter **I**nheri**T**ance **H**yperNetwork (**PITH**), which introduces a novel parameter projection mechanism to directly inherit parameters from pre-trained models for initializing target networks of varying configurations. Our method enables initialized networks to directly achieve competitive performance on downstream tasks without any further training, which we term zero-shot initialization. Extensive experiments demonstrate the superiority of PITH: ViT-Base initialized by PITH achieves 53.35% zero-shot accuracy on ImageNet-1K, surpassing the previous state-of-the-art by 6.54%, with consistent improvements across multiple downstream tasks.

Diffusion probabilistic models have demonstrated significant potential in generating high-quality, realistic medical images, providing a promising solution to the persistent challenge of data scarcity in the medical field. Nevertheless, producing 3D medical volumes with anatomically consistent structures under multimodal conditions remains a complex and unresolved problem. We introduce Sketch2CT, a multimodal diffusion framework for structure-aware 3D medical volume generation, jointly guided by a user-provided 2D sketch and a textual description that captures 3D geometric semantics. The framework initially generates 3D segmentation masks of the target organ from random noise, conditioned on both modalities. To effectively align and fuse these inputs, we propose two key modules that refine sketch features with localized textual cues and integrate global sketch-text representations. Built upon a capsule-attention backbone, these modules leverage the complementary strengths of sketches and text to produce anatomically accurate organ shapes. The synthesized segmentation masks subsequently guide a latent diffusion model for 3D CT volume synthesis, enabling realistic reconstruction of organ appearances that are consistent with user-defined sketches and descriptions. Extensive experiments on public CT datasets demonstrate that Sketch2CT achieves superior performance in generating multimodal medical volumes. Its controllable, low-cost generation pipeline enables principled, efficient augmentation of medical datasets. Code is available at https://github.com/adlsn/Sketch2CT.

End-to-end multi-object tracking (MOT) methods have recently achieved remarkable progress by unifying detection and association within a single framework. Despite their strong detection performance, these methods suffer from relatively low association accuracy. Through detailed analysis, we observe that object embeddings produced by the shared DETR architecture display excessively high inter-object similarity, as it emphasizes only category-level discrimination within single frames. In contrast, tracking requires instance-level distinction across frames with spatial and temporal continuity, for which current end-to-end approaches insufficiently optimize object embeddings. To address this, we introduce FDTA (From Detection to Association), an explicit feature refinement framework that enhances object discriminativeness across three complementary perspectives. Specifically, we introduce a Spatial Adapter (SA) to integrate depth-aware cues for spatial continuity, a Temporal Adapter (TA) to aggregate historical information for temporal dependencies, and an Identity Adapter (IA) to leverage quality-aware contrastive learning for instance-level separability. Extensive experiments demonstrate that FDTA achieves state-of-the-art performance on multiple challenging MOT benchmarks including DanceTrack, SportsMOT, and BFT, highlighting the effectiveness of our proposed discriminative embedding enhancement strategy. The code is available at https://github.com/Spongebobbbbbbbbbb/FDTA.

Large Vision Language Models (LVLMs) have demonstrated remarkable capabilities in Chain-of-Thought (CoT) reasoning. However, existing LVLM reasoning paradigms only begin reasoning after the entire video becomes available, introducing unnecessary latency and diminishing attention to early visual cues in dynamic scenes. Inspired by the human ability to think while watching, we introduce a streaming reasoning paradigm for LVLMs, where reasoning unfolds sequentially with incoming frames and deepens after the full video is observed. We instantiate this paradigm through Think-as-You-See (TaYS), a unified framework that enables LVLMs to reason while watching by integrating streaming CoT generation, stream-constrained training, and stream-parallel inference. Specifically, TaYS employs temporally aligned streaming reasoning units with precise CoT supervision, enforces ordered reasoning via streaming attention masks and positional encodings, and utilizes a parallel KV caches mechanism that decouples input encoding from reasoning generation, ensuring alignment and true concurrency. We evaluate TaYS on the Qwen2.5-VL model family across representative video CoT tasks, including event dynamics analysis, causal reasoning, and thematic understanding. Experimental results show that TaYS achieves superior reasoning performance compared with batch-mode CoT, while reducing pre-reasoning latency to under one second and overall answer delay by more than 50%. These findings demonstrate the effectiveness of the streaming paradigm in enabling real-time, human-like reasoning for LVLMs.