ICCV 2025 论文列表

Text-to-image diffusion models have seen significant development recently due to increasing availability of paired 2D data. Although a similar trend is emerging in 3D generation, the limited availability of high-quality 3D data has resulted in less competitive 3D diffusion models compared to their 2D counterparts. In this work, we show how 2D diffusion models, originally trained for text-to-image generation, can be repurposed for 3D object generation. We introduce Gaussian Atlas, a representation of 3D Gaussians with dense 2D grids, which enables the fine-tuning of 2D diffusion models for generating 3D Gaussians. Our approach shows a successful transfer learning from a pretrained 2D diffusion model to 2D manifold flattend from 3D structures. To facilitate model training, a large-scale dataset, Gaussian Atlas, is compiled to comprise 205K high-quality 3D Gaussian fittings of a diverse array of 3D objects. Our experiment results indicate that text-to-image diffusion models can also serve as 3D content generators.

Constructing a compressed latent space through a variational autoencoder (VAE) is the key for efficient 3D diffusion models. This paper introduces COD-VAE that encodes 3D shapes into a COmpact set of 1D latent vectors without sacrificing quality. COD-VAE introduces a two-stage autoencoder scheme to improve compression and decoding efficiency. First, our encoder block progressively compresses point clouds into compact latent vectors via intermediate point patches. Second, our triplane-based decoder reconstructs dense triplanes from latent vectors instead of directly decoding neural fields, significantly reducing computational overhead of neural fields decoding. Finally, we propose uncertainty-guided token pruning, which allocates resources adaptively by skipping computations in simpler regions and improves the decoder efficiency. Experimental results demonstrate that COD-VAE achieves 16x compression compared to the baseline while maintaining quality. This enables 20.8x speedup in generation, highlighting that a large number of latent vectors is not a prerequisite for high-quality reconstruction and generation.

Transformers have demonstrated remarkable success across vision, language, and video. Yet, increasing task complexity has led to larger models and more tokens, raising the quadratic cost of self-attention and the overhead of GPU memory access. To reduce the computation cost of self-attention, prior work has proposed token compression techniques that drop redundant or less informative tokens. Meanwhile, fused attention kernels such as FlashAttention have been developed to alleviate memory overhead by avoiding attention map construction and its associated I/O to HBM. This, however, makes it incompatible with most training-free token compression methods, which rely on attention maps to determine token importance. Here, we propose Representation Shift, a training-free, model-agnostic metric that measures the degree of change in each token's representation. This seamlessly integrates token compression with FlashAttention, without attention maps or retraining. Our method further generalizes beyond Transformers to CNNs and state space models. Extensive experiments show that Representation Shift enables effective token compression compatible with FlashAttention, yielding significant speedups of up to 5.5x and 4.4x in video-text retrieval and video QA, respectively. Code is available at https://github.com/mlvlab/Representation-Shift.

Despite the promise of Multi-Task Learning (MTL) in leveraging complementary knowledge across tasks, existing multi-task optimization (MTO) techniques remain fixated on resolving conflicts through optimizer-centric loss scaling and gradient manipulation, yet fail to deliver consistent gains. In this paper, we argue that the shared representation space, where task interactions naturally occur, offers rich information and potential for operations complementary to existing optimizer designs, especially for facilitating the inter-task complementarity, which is rarely explored in MTO. This intuition leads to Rep-MTL, which exploits the representation-level task saliency to quantify interactions between task-specific optimization and shared representation learning. By steering these saliencies through entropybased penalization and sample-wise cross-task alignment, Rep-MTL aims to mitigate negative transfer by maintaining the effective training of individual tasks instead pure conflict-solving, while explicitly promoting complementary information sharing. Experiments are conducted on four challenging MTL benchmarks covering both task-shift and domain-shift scenarios. The results show that Rep-MTL, even paired with the basic equal weighting (EW) policy, achieves competitive performance gains with favorable efficiency. Beyond standard performance metrics, Power Law (PL) exponent analysis demonstrates Rep-MTL's efficacy in balancing task-specific learning and cross-task sharing.

Reflective flares are common artifacts in photography that degrade image quality, introducing in-focus flares, which appear as bright, regular spot patterns, and out-of-focus flares, which are diffuse and semi-transparent, obscuring the underlying scene. While previous methods have achieved some success in removing in-focus flares, they struggle with the diffuse nature of out-of-focus flares. The lack of an out-of-focus flare dataset has further hindered the development of effective flare removal models. In this work, we construct a large-scale out-of-focus flare dataset generated with Blender. We propose to use a color alignment approach on diffusion models to address the challenges of out-of-focus reflective flare removal. Rather than fully reconstructing flare-affected regions, our method adjusts the color distribution to reduce artifact visibility while preserving image content. Specifically, we apply a differentiable histogram loss on the diffusion model, which is derived from the Earth Mover's Distance (EMD), to effectively align color distributions. The proposed approach, trained exclusively on synthetic data, outperforms existing methods on both synthetic and real-world data, demonstrating improved performance in flare removal.

Cost volumes are used in every modern optical flow estimator, but due to their computational and space complexity, they are often a limiting factor regarding both processing speed and the resolution of input frames. Motivated by our empirical observation that cost volumes lose their importance once all other network parts of, e.g., a RAFT-based pipeline have been sufficiently trained, we introduce a training strategy that allows removing the cost volume from optical flow estimators throughout training. This leads to significantly improved inference speed and reduced memory requirements. Using our training strategy, we create three different models covering different compute budgets. Our most accurate model reaches state-of-the-art accuracy while being 1.2xfaster and having a 6xlower memory footprint than comparable models; our fastest model is capable of processing Full HD frames at 20\,\mathrm FPS using only 500\,\mathrm MB of GPU memory.

Machine unlearning enables the removal of specific data from ML models to uphold the *right to be forgotten*. While approximate unlearning algorithms offer efficient alternatives to full retraining, this work reveals that they fail to adequately protect the privacy of unlearned data. In particular, these algorithms introduce implicit residuals which facilitate privacy attacks targeting at unlearned data. We observe that these residuals persist regardless of model architectures, parameters, and unlearning algorithms, exposing a new attack surface beyond conventional output-based leakage. Based on this insight, we propose the *Reminiscence Attack (ReA)*, which amplifies the correlation between residuals and membership privacy through targeted fine-tuning processes. ReA achieves up to 1.90x and 1.12x higher accuracy than prior attacks when inferring class-wise and sample-wise membership, respectively. To mitigate such residual-induced privacy risk, we develop a dual-phase approximate unlearning framework that first eliminates deep-layer unlearned data traces and then enforces convergence stability to prevent models from "pseudo-convergence", where their outputs are similar to retrained models but still preserve unlearned residuals. Our framework works for both classification and generation tasks. Experimental evaluations confirm that our approach maintains high unlearning efficacy, while reducing the adaptive privacy attack accuracy to nearly random guess, at the computational cost of 2-12% of full retraining.

We address the challenge of constructing a consistent and photorealistic Neural Radiance Field in inhomogeneously illuminated, scattering environments with unknown, co-moving light sources. While most existing works on underwater scene representation focus on a static homogeneous illumination, limited attention has been paid to scenarios such as when a robot explores water deeper than a few tens of meters, where sunlight becomes insufficient. To address this, we propose a novel illumination field locally attached to the camera, enabling the capture of uneven lighting effects within the viewing frustum. We combine this with a volumetric medium representation to an overall method that effectively handles interaction between dynamic illumination field and static scattering medium. Evaluation results demonstrate the effectiveness and flexibility of our approach.

Modern deep architectures often rely on large-scale datasets, but training on these datasets incurs high computational and storage overhead. Real-world datasets often contain substantial redundancies, prompting the need for more data-efficient training paradigms. Data selection has shown promise to mitigate redundancy by identifying the most representative samples, thereby reducing training costs without compromising performance. Existing methods typically rely on static scoring metrics or pretrained models, overlooking the combined effect of selected samples and their evolving dynamics during training. To address this, we introduce the concept of epsilon-sample cover, which quantifies sample redundancy based on inter-sample relationships, capturing the intrinsic structure of the dataset. Based on this, we reformulate data selection as a reinforcement learning (RL) process, where a lightweight RL agent optimizes the selection policy by leveraging epsilon-sample cover derived from evolving dataset distribution as a reward signal. Extensive experiments across benchmark datasets and diverse architectures demonstrate that our method consistently outperforms existing state-of-the-art baselines. Models trained with our selected datasets show enhanced generalization performance with improved training efficiency.

Affine Grassmannian has been favored for expressing proximity between lines and planes due to its theoretical exactness in measuring distances among features. Despite this advantage, the existing method can only measure the proximity without yielding the distance as an explicit function of rigid body transformation. Thus, an optimizable distance function on the manifold has remained underdeveloped, stifling its application in registration problems. This paper is the first to explicitly derive an optimizable cost function between two Grassmannian features with respect to rigid body transformation (R and t). Specifically, we present a rigorous mathematical proof demonstrating that the bases of high-dimensional linear subspaces can serve as an explicit representation of the cost. Finally, we propose an optimizable cost function based on the transformed bases that can be applied to the registration problem of any affine subspace. Compared to vector parameter-based approaches, our method is able to find a globally optimal solution by directly minimizing the geodesic distance which is agnostic to representation ambiguity. The resulting cost function and its extension to the inlier-set maximizing Branch-and-Bound (BnB) solver have been demonstrated to improve the convergence of existing solutions or outperform them in various computer vision tasks. The code is available on https://github.com/joomeok/GrassmannRegistration.

Learning visual representations is foundational for a broad spectrum of downstream tasks. Although recent vision-language contrastive models, such as CLIP and SigLIP, have achieved impressive zero-shot performance via large-scale vision-language alignment, their reliance on global representations constrains their effectiveness for dense prediction tasks, such as grounding, OCR, and segmentation. To address this gap, we introduce Region-Aware Cluster Discrimination (RICE), a novel method that enhances region-level visual and OCR capabilities. We first construct a billion-scale candidate region dataset and propose a Region Transformer layer to extract rich regional semantics. We further design a unified region cluster discrimination loss that jointly supports object and OCR learning within a single classification framework, enabling efficient and scalable distributed training on large-scale data. Extensive experiments show that RICE consistently outperforms previous methods on tasks, including segmentation, dense detection, and visual perception for Multimodal Large Language Models (MLLMs). The pre-trained models have been released at https://github.com/deepglint/MVT.

Referring Visual Grounding (RVG) tasks revolve around utilizing vision-language interactions to incorporate object information from language expressions, thereby enabling targeted object detection or segmentation within images. Transformer-based methods have enabled effective interaction through attention mechanisms, achieving notable performance in RVG tasks. However, existing strategies for RVG, which involve direct interaction between visual and linguistic features, face three key challenges: (i) tendency to focus on a single target, (ii) insufficient control over linguistic noise, and (iii) high computational cost. To address these challenges, we propose a Region-aware Anchoring Mechanism (RaAM) that mediates vision-language interactions. In RaAM, region-aware anchors engage in alternating interactions with vision and language modalities, acting as indicators for object presence across different regions within the image. RaAM (i) directs attention to multiple target regions for better localization, (ii) reduces cross-modal redundancy by using anchors as buffers, and (iii) lowers time complexity. In addition, we design region and pixel level loss functions to enhance object presence assessment and edge precision. We evaluate our RaAM-RVG on four benchmark datasets and integrate RaAM into various models by replacing their interaction design. Results show that RaAM outperforms state-of-the-art methods with lower computational cost.

Data attribution in text-to-image generative models is a crucial yet underexplored problem, particularly at the regional level, where identifying the most influential training regions for generated content can enhance transparency, copyright protection, and error diagnosis. Existing data attribution methods either operate at the whole-image level or lack scalability for large-scale generative models. In this work, we propose a novel framework for region-level data attribution. At its core is the Attribution Region (AR) detector, which localizes influential regions in training images used by the text-to-image generative model. To support this research, we construct a large-scale synthetic dataset with ground-truth region-level attributions, enabling both training and evaluation of our method. Empirical results show that our method outperforms existing attribution techniques in accurately tracing generated content back to training data. Additionally, we demonstrate practical applications, including identifying artifacts in generated images and suggesting improved replacements for generated content. Our dataset and framework will be released to advance further research in region-level data attribution for generative models.

3D Gaussian Splatting (3DGS) has demonstrated its potential in reconstructing scenes from unposed images. However, optimization-based 3DGS methods struggle with sparse views due to limited prior knowledge. Meanwhile, feed-forward Gaussian approaches are constrained by input formats, making it challenging to incorporate more input views. To address these challenges, we propose RegGS, a 3D Gaussian registration-based framework for reconstructing unposed sparse views. RegGS aligns local 3D Gaussians generated by a feed-forward network into a globally consistent 3D Gaussian representation. Technically, we implement an entropy-regularized Sinkhorn algorithm to efficiently solve the optimal transport Mixture 2-Wasserstein MW_2 distance, which serves as an alignment metric for Gaussian mixture models (GMMs) in Sim(3) space. Furthermore, we design a joint 3DGS registration module that integrates the MW_2 distance, photometric consistency, and depth geometry. This enables a coarse-to-fine registration process while accurately estimating camera poses and aligning the scene. Experiments on the RE10K and ACID datasets demonstrate that RegGS effectively registers local Gaussians with high fidelity, achieving precise pose estimation and high-quality novel view synthesis.

The predominant approach to advancing text-to-image generation has been training-time scaling, where larger models are trained on more data using greater computational resources. While effective, this approach is computationally expensive, leading to growing interest in inference-time scaling to improve performance. Currently, inference-time scaling for text-to-image diffusion models is largely limited to best-of-N sampling, where multiple images are generated per prompt and a selection model chooses the best output. Inspired by the recent success of reasoning models like DeepSeek-R1 in the language domain, we introduce an alternative to naive best-of-N sampling by equipping text-to-image Diffusion Transformers with in-context reflection capabilities. We propose Reflect-DiT, a method that enables Diffusion Transformers to refine their generations using in-context examples of previously generated images alongside textual feedback describing necessary improvements. Instead of passively relying on random sampling and hoping for a better result in a future generation, Reflect-DiT explicitly tailors its generations to address specific aspects requiring enhancement. Experimental results demonstrate that Reflect-DiT improves performance on the GenEval benchmark (+0.19) using SANA-1.0-1.6B as a base model. Additionally, it achieves a new state-of-the-art score of 0.81 on GenEval while generating only 20 samples per prompt, surpassing the previous best score of 0.80, which was obtained using a significantly larger model (SANA-1.5-4.8B) with 2048 samples under the best-of-N approach.

Humans are undoubtedly the most important participants in computer vision, and the ability to detect any individual given a natural language description, a task we define as referring to any person, holds substantial practical value. However, we find that existing models generally fail to achieve real-world usability, and current benchmarks are limited by their focus on one-to-one referring, that hinder progress in this area. In this work, we revisit this task from three critical perspectives: task definition, dataset design, and model architecture. We first identify five aspects of referable entities and three distinctive characteristics of this task. Next, we introduce HumanRef, a novel dataset designed to tackle these challenges and better reflect real-world applications. From a model design perspective, we integrate a multimodal large language model with an object detection framework, constructing a robust referring model named RexSeek. Experimental results reveal that state-of-the-art models, which perform well on commonly used benchmarks like RefCOCO/+/g, struggle with HumanRef due to their inability to detect multiple individuals. In contrast, RexSeek not only excels in human referring but also generalizes effectively to common object referring, making it broadly applicable across various perception tasks.

Referring expression comprehension (REC) aims to localize the target object described by a natural language expression.Recent advances in vision-language learning have led to significant performance improvements in REC tasks.However, localizing extremely small objects remains a considerable challenge despite its importance in real-world applications such as autonomous driving.To address this issue, we introduce a novel dataset and method for REC targeting small objects.First, we present the small object REC (SOREC) dataset, which consists of 100,000 pairs of referring expressions and corresponding bounding boxes for small objects in driving scenarios.Second, we propose the progressive-iterative zooming adapter (PIZA), an adapter module for parameter-efficient fine-tuning that enables models to progressively zoom in and localize small objects.In a series of experiments, we apply PIZA to GroundingDINO and demonstrate a significant improvement in accuracy on the SOREC dataset.Our dataset, codes and pre-trained models are publicly available on the project page.

Most existing diffusion models have primarily utilized reference images for image-to-image translation rather than for super-resolution (SR). In SR-specific tasks, diffusion methods rely solely on low-resolution (LR) inputs, limiting their ability to leverage reference information. Prior reference-based diffusion SR methods have shown that incorporating appropriate references can significantly enhance reconstruction quality; however, identifying suitable references in real-world scenarios remains a critical challenge. Recently, Retrieval-Augmented Generation (RAG) has emerged as an effective framework that integrates retrieval-based and generation-based information from databases to enhance the accuracy and relevance of responses. Inspired by RAG, we propose an image-based RAG framework (iRAG) for realistic super-resolution, which employs a trainable hashing function to retrieve either real-world or generated references given an LR query. Retrieved patches are passed to a restoration module that generates high-fidelity super-resolved features, and a hallucination filtering mechanism is used to refine generated references from pre-trained diffusion models. Experimental results demonstrate that our approach not only resolves practical difficulties in reference selection but also delivers superior performance over existing diffusion and non-diffusion RefSR methods. Code is available at https://github.com/ByeonghunLee12/iRAG.

We present REM, a framework for segmenting a wide range of concepts in video that can be described through natural language. Our method leverages the universal visual-language mapping learned by video diffusion models on Internet-scale data by fine-tuning them on small-scale Referring Object Segmentation datasets. Our key insight is to preserve the entirety of the generative model's architecture by shifting its objective from predicting noise to predicting mask latents. The resulting model can accurately segment rare and unseen objects, despite only being trained on a limited set of categories. Additionally, it can effortlessly generalize to non-object dynamic concepts, such as smoke or raindrops, as demonstrated in our new benchmark for Referring Video Process Segmentation (Ref-VPS). REM performs on par with the state-of-the-art on in-domain datasets, like Ref-DAVIS, while outperforming them by up to 12 IoU points out-of-domain, leveraging the power of generative pre-training. We also show that advancements in video generation directly improve segmentation.