ICCV 2025 Papers

Humans perceive the world through multimodal cues to understand and interact with the environment. Learning a scene representation for multiple modalities enhances comprehension of the physical world. However, modality conflicts, arising from inherent distinctions among different modalities, present two critical challenges: property disparity and granularity disparity. To address these challenges, we propose a general framework, MMOne, to represent multiple modalities in one scene, which can be readily extended to additional modalities. Specifically, a modality modeling module with a novel modality indicator is proposed to capture the unique properties of each modality. Additionally, we design a multimodal decomposition mechanism to separate multi-modal Gaussians into single-modal Gaussians based on modality differences. We address the essential distinctions among modalities by disentangling multimodal information into shared and modality-specific components, resulting in a more compact and efficient multimodal scene representation. Extensive experiments demonstrate that our method consistently enhances the representation capability for each modality and is scalable to additional modalities. The code is available at https://github.com/Neal2020GitHub/MMOne.

Multimodal geo-localization methods can inherently overcome the limitations of unimodal sensor systems by leveraging complementary information from different modalities. However, existing retrieval-based methods rely on a comprehensive multimodal database, which is often challenging to fulfill in practice. In this paper, we introduce a more practical problem for localizing drone-view images by collaborating multimodal data within a satellite-view reference map, which integrates multimodal information while avoiding the need for an extensive multimodal database. We present MMGeo that learns to push the composition of multimodal representations to the target reference map through a unified framework. By utilizing a comprehensive multimodal query (image, point cloud/depth/text), we can achieve more robust and accurate geo-localization, especially in unknown and complex environments. Additionally, we extend two visual geo-localization datasets GTA-UAV and UAV-VisLoc to multi-modality, establishing the first UAV geo-localization datasets that combine image, point cloud, depth and text data. Experiments demonstrate the effectiveness of MMGeo for UAV multimodal compositional geo-localization, as well as the generalization capabilities to real-world scenarios. The code and dataset are at https://github.com/Yux1angJi/MMGeo.

Fully comprehending scientific papers by machines reflects a high level of Artificial General Intelligence, requiring the ability to reason across fragmented and heterogeneous sources of information, presenting a complex and practically significant challenge. While Vision-Language Models (VLMs) have made remarkable strides in various tasks, particularly those involving reasoning with evidence source from single image or text page, their ability to use cross-source information for reasoning remains an open problem. This work presents MMCR, a high-difficulty benchmark designed to evaluate VLMs' capacity for reasoning with cross-source information from scientific papers. The benchmark comprises 276 high-quality questions, meticulously annotated by humans across 7 subjects and 10 task types. Experiments with 18 VLMs demonstrate that cross-source reasoning presents a substantial challenge for existing models. Notably, even the top-performing model, GPT-4o, achieved only 48.55% overall accuracy, with just 20% accuracy in multi-table comprehension tasks, while the second-best model, Qwen2.5-VL-72B, reached 39.86% overall accuracy. These results highlight the pressing need to develop VLMs capable of effectively utilizing cross-source information for reasoning.

Large Language Models (LLMs), enhanced through agent tuning, have demonstrated remarkable capabilities in Chain-of-Thought (CoT) and tool utilization, significantly surpassing the performance of standalone models. However, the multimodal domain still lacks a large-scale, high-quality agent tuning dataset to unlock the full potential of multimodal large language models. To bridge this gap, we introduce MMAT-1M, the first million-scale multimodal agent tuning dataset designed to support CoT, reflection, and dynamic tool usage. Our dataset is constructed through a novel four-stage data engine: 1) We first curate publicly available multimodal datasets containing question-answer pairs; 2) Then, leveraging GPT-4o, we generate rationales for the original question-answer pairs and dynamically integrate API calls and Retrieval Augmented Generation (RAG) information through a multi-turn paradigm; 3) Furthermore, we refine the rationales through reflection to ensure logical consistency and accuracy, creating a multi-turn dialogue dataset with both Rationale and Reflection (RR); 4) Finally, to enhance efficiency, we optionally compress multi-turn dialogues into a One-turn Rationale and Reflection (ORR) format. By fine-tuning open-source multimodal models on the MMAT-1M, we observe significant performance gains. For instance, the InternVL2.5-8B-RR model achieves an average improvement of 2.7% across eight public benchmarks and 8.8% on the RAG benchmark Dyn-VQA, demonstrating the dataset's effectiveness in enhancing multimodal reasoning and tool-based capabilities. The dataset is publicly available at https://github.com/VIS-MPU-Agent/MMAT-1M.

Image fusion, a fundamental low-level vision task, aims to integrate multiple image sequences into a single output while preserving as much information as possible from the input. However, existing methods face several significant limitations: 1) requiring task- or dataset-specific models; 2) neglecting real-world image degradations (e.g., noise), which causes failure when processing degraded inputs; 3) operating in pixel space, where attention mechanisms are computationally expensive; and 4) lacking user interaction capabilities.To address these challenges, we propose a unified framework for multi-task, multi-degradation, and language-guided image fusion. Our framework includes two key components: 1) a practical degradation pipeline that simulates real-world image degradations and generates interactive prompts to guide the model; 2) an all-in-one Diffusion Transformer (DiT) operating in latent space, which fuses a clean image conditioned on both the degraded inputs and the generated prompts. Furthermore, we introduce principled modifications to the original DiT architecture to better suit the fusion task. Based on this framework, we develop two versions of the model: Regression-based and Flow Matching-based variants.Extensive qualitative and quantitative experiments demonstrate that our approach effectively addresses the aforementioned limitations and outperforms previous restoration+fusion and all-in-one pipelines.

Human body actions are an important form of non-verbal communication in social interactions. This paper specifically focuses on a subset of body actions known as micro-actions, which are subtle, low-intensity body movements with promising applications in human emotion analysis. In real-world scenarios, human micro-actions often temporally co-occur, with multiple micro-actions overlapping in time, such as concurrent head and hand movements. However, current research primarily focuses on recognizing individual micro-actions while overlooking their co-occurring nature. To address this gap, we propose a new task named Multi-label Micro-Action Detection (MMAD), which involves identifying all micro-actions in a given short video, determining their start and end times, and categorizing them. Accomplishing this requires a model capable of accurately capturing both long-term and short-term action relationships to detect multiple overlapping micro-actions. To facilitate the MMAD task, we introduce a new dataset named Multi-label Micro-Action-52 (MMA-52) and propose a baseline method equipped with a dual-path spatial-temporal adapter to address the challenges of subtle visual change in MMAD. We hope that MMA-52 can stimulate research on micro-action analysis in videos and prompt the development of spatio-temporal modeling in human-centric video understanding. The proposed MMA-52 dataset is available at: https://github.com/VUT-HFUT/Micro-Action

Multimodal large language models (MLLMs) excel at 2D visual understanding but remain limited in their ability to reason about 3D space. In this work, we leverage large-scale high-quality 3D scene data with open-set annotations to introduce 1) a novel supervised fine-tuning dataset and 2) a new evaluation benchmark, focused on indoor scenes. Our Cubify Anything VQA (CA-VQA) data covers diverse spatial tasks including spatial relationship prediction, metric size and distance estimation, and 3D grounding. We show that CA-VQA enables us to train MM-Spatial, a strong generalist MLLM that also achieves state-of-the-art performance on 3D spatial understanding benchmarks, including our own. We show how incorporating metric depth and multi-view inputs (provided in CA-VQA) can further improve 3D understanding, and demonstrate that data alone allows our model to achieve depth perception capabilities comparable to dedicated monocular depth estimation models.

The Instruction Following (IF) ability measures how well Multi-modal Large Language Models (MLLMs) understand exactly what users are telling them and doing it right.Existing multimodal instruction following training data is scarce, the benchmarks are simple with atomic instructions, and the evaluation strategies are imprecise for tasks demanding exact output constraints.To address this, we present MM-IFEngine, an effective pipeline to generate high-quality image-instruction pairs.Our MM-IFEngine pipeline yields large-scale, diverse, and high-quality training data MM-IFInstruct-23k, which is suitable for Supervised Fine-Tuning (SFT) and extended as MM-IFDPO-23k for Direct Preference Optimization (DPO).We further introduce MM-IFEval, a challenging and diverse multi-modal instruction-following benchmark that includes (1) both textual constraints for output responses and visual constraints tied to the input images, and (2) a comprehensive evaluation pipeline incorporating rule-based assessment and LLM-as-a-Judge evaluation.We conduct SFT and DPO experiments and demonstrate that fine-tuning MLLMs on MM-IFInstruct-23k and MM-IFDPO-23k achieve notable gains on various IF benchmarks, such as MM-IFEval (+11.8%), MIA (+7.7%), and IFEval (+10.5%).

We introduce MIORe and VAR-MIORe, two novel multi-task datasets that address critical limitations in current motion restoration benchmarks. Designed with high-frame-rate (1000 FPS) acquisition and professional-grade optics, our datasets capture a broad spectrum of motion scenarios, which include complex ego-camera movements, dynamic multi-subject interactions, and depth-dependent blur effects. By adaptively averaging frames based on computed optical flow metrics, MIORe generates consistent motion blur, and preserves sharp inputs for video frame interpolation and optical flow estimation. VAR-MIORe further extends by spanning a variable range of motion magnitudes, from minimal to extreme, establishing the first benchmark to offer explicit control over motion amplitude. We provide high-resolution, scalable ground truths that challenge existing algorithms under both controlled and adverse conditions, paving the way for next-generation research of various image and video restoration tasks.

Multimodal LLMs are turning their focus to video benchmarks, however most video benchmarks only provide outcome supervision, with no intermediate or interpretable reasoning steps. This makes it challenging to assess if models are truly able to combine perceptual and temporal information to reason about videos, or simply get the correct answer by chance or by exploiting linguistic biases. To remedy this, we provide a new video reasoning dataset called MINERVA for modern multimodal models. Each question in the dataset comes with 5 answer choices, as well as detailed, hand-crafted reasoning traces. Our dataset is multimodal, diverse in terms of video domain and length, and consists of complex multi-step questions. Extensive benchmarking shows that our dataset provides a challenge for frontier open-source and proprietary models. We perform fine-grained error analysis to identify common failure modes across various models, and create a taxonomy of reasoning errors. We use this to explore both human and LLM-as-a-judge methods for scoring video reasoning traces, and find that failure modes are primarily related to temporal localization, followed by visual perception errors, as opposed to logical or completeness errors. The dataset, along with questions, answer candidates and reasoning traces is publicly available under https://github.com/google-deepmind/neptune?tab=readme-ov-file#minerva https://github.com/google-deepmind/neptune?tab=readme-ov-file\#minerva.

Image representations are often evaluated through disjointed, task-specific protocols, leading to a fragmented understanding of model capabilities. For instance, it is unclear whether an image embedding model adept at clustering images is equally good at retrieving relevant images given a piece of text. We introduce the Massive Image Embedding Benchmark (MIEB) to evaluate the performance of image and image-text embedding models across the broadest spectrum to date. MIEB spans 38 languages across 130 individual tasks, which we group into 8 high-level categories. We benchmark 50 models across our benchmark, finding that no single method dominates across all task categories. We reveal hidden capabilities in advanced vision models such as their accurate visual representation of texts, and their yet limited capabilities in interleaved encodings and matching images and texts in the presence of confounders. We also show that the performance of vision encoders on MIEB correlates highly with their performance when used in multimodal large language models. Our code, dataset, and leaderboard are publicly available at https://github.com/embeddings-benchmark/mteb.

This work, termed MH-LVC, presents a multi-hypothesis temporal prediction scheme that employs long- and short-term reference frames in a conditional residual video coding framework. Recent temporal context mining approaches to conditional video coding offer superior coding performance. However, the need to store and access a large amount of implicit contextual information extracted from past decoded frames in decoding a video frame poses a challenge due to excessive memory access. Our MH-LVC overcomes this issue by storing multiple long- and short-term reference frames but limiting the number of reference frames used at a time for temporal prediction to two. Our decoded frame buffer management allows the encoder to flexibly utilize the long-term key frames to mitigate temporal cascading errors and the short-term reference frames to minimize prediction errors. Moreover, our buffering scheme enables the temporal prediction structure to be adapted to individual input videos. While this flexibility is common in traditional video codecs, it has not been fully explored for learned video codecs. Extensive experiments show that the proposed method outperforms VTM-17.0 under the low-delay B configuration in terms of PSNR-RGB across commonly used test datasets, and performs comparably to the state-of-the-art learned codecs (e.g. DCVC-FM) while requiring less decoded frame buffer and similar decoding time.

Multi-camera systems are increasingly vital in the environmental perception of autonomous vehicles and robotics. Their physical configuration offers inherent fixed relative pose constraints that benefit Structure-from-Motion (SfM). However, traditional global SfM systems struggle with robustness due to their optimization framework.We propose a novel global motion averaging framework for multi-camera systems, featuring two core components: a decoupled rotation averaging module and a hybrid translation averaging module.Our rotation averaging employs a hierarchical strategy by first estimating relative rotations within rigid camera units and then computing global rigid unit rotations.To enhance the robustness of translation averaging, we incorporate both camera-to-camera and camera-to-point constraints to initialize camera positions and 3D points with a convex distance-based objective function and refine them with an unbiased non-bilinear angle-based objective function.Experiments on large-scale datasets show that our system matches or exceeds incremental SfM accuracy while significantly improving efficiency.Our framework outperforms existing global SfM methods, establishing itself as a robust solution for real-world multi-camera SfM applications. We will share our system as an open-source implementation.

Novel view synthesis (NVS) and surface reconstruction (SR) are essential tasks in 3D Gaussian Splatting (3DGS). Despite recent progress, these tasks are often addressed independently, with GS-based rendering methods struggling under diverse light conditions and failing to produce accurate surfaces, while GS-based reconstruction methods frequently compromise rendering quality. This raises a central question: must rendering and reconstruction always involve a trade-off? To address this, we propose MGSR, a 2D/3D Mutual-boosted Gaussian Splatting for Surface Reconstruction that enhances both rendering quality and 3D reconstruction accuracy. MGSR introduces two branches--one based on 2DGS and the other on 3DGS. The 2DGS branch excels in surface reconstruction, providing precise geometry information to the 3DGS branch. Leveraging this geometry, the 3DGS branch employs a geometry-guided illumination decomposition module that captures reflected and transmitted components, enabling realistic rendering under varied light conditions. Using the transmitted component as supervision, the 2DGS branch also achieves high-fidelity surface reconstruction. Throughout the optimization process, the 2DGS and 3DGS branches undergo alternating optimization, providing mutual supervision. Prior to this, each branch completes an independent warm-up phase, with an early stopping strategy implemented to reduce computational costs. We evaluate MGSR on a diverse set of synthetic and real-world datasets, at both object and scene levels, demonstrating strong performance in rendering and surface reconstruction. Code is available at https://github.com/TsingyuanChou/MGSR.

Vision encoders serve as the cornerstone of multimodal understanding. Single-encoder architectures like CLIP exhibit inherent constraints in generalizing across diverse multimodal tasks, while recent multi-encoder fusion methods introduce prohibitive computational overhead to achieve superior performance using complementary visual representations from multiple vision encoders. To address this, we propose a progressive pruning framework, namely Multi-Encoder collaboraTivE tOken pRuning (METEOR), that eliminates redundant visual tokens across the encoding, fusion, and decoding stages for multi-encoder MLLMs. For multi-vision encoding, we discard redundant tokens within each encoder via a rank guided collaborative token assignment strategy. Subsequently, for multi-vision fusion, we combine the visual features from different encoders while reducing cross-encoder redundancy with cooperative pruning. Finally, we propose an adaptive token pruning method in the LLM decoding stage to further discard irrelevant tokens based on the text prompts with dynamically adjusting pruning ratios for specific task demands. To our best knowledge, this is the first successful attempt that achieves an efficient multi-encoder based vision language model with multi-stage pruning strategies. Extensive experiments on 11 benchmarks demonstrate the effectiveness of our proposed approach. Compared with EAGLE, a typical multi-encoder MLLMs, METEOR reduces 76% visual tokens with only 0.3% performance drop in average. The code is available at https://github.com/YuchenLiu98/METEOR.

Recent advances in optical flow estimation have prioritized accuracy at the cost of growing GPU memory consumption, particularly for high-resolution (FullHD) inputs. We introduce MEMFOF, a memory-efficient multi-frame optical flow method that identifies a favorable trade-off between multi-frame estimation and GPU memory usage. Notably, MEMFOF requires only 2.09 GB of GPU memory at runtime for 1080p inputs, and 28.5 GB during training, which uniquely positions our method to be trained at native 1080p without the need for cropping or downsampling. We systematically revisit design choices from RAFT-like architectures, integrating reduced correlation volumes and high-resolution training protocols alongside multi-frame estimation, to achieve state-of-the-art performance across multiple benchmarks while substantially reducing memory overhead. Our method outperforms more resource-intensive alternatives in both accuracy and runtime efficiency, validating its robustness for flow estimation at high resolutions. At the time of submission, our method ranks first on the Spring benchmark with a 1-pixel (1px) outlier rate of 3.289, leads Sintel (clean) with an endpoint error (EPE) of 0.963, and achieves the best Fl-all error on KITTI-2015 at 2.94%. The code is available at: https://github.com/msu-video-group/memfof.

The recognition of Ancient Egyptian hieroglyphs poses persistent challenges due to limited annotated data and wide stylistic variation. We introduce the Multisource Egyptian Hieroglyphs (MEH) dataset, a new benchmark that captures a diverse range of writing styles with detailed clause-level and OCR annotations derived from expert-curated sources. To address the scarcity of training data, we propose a method for generating synthetic hieroglyphic sequences by combining real linguistic material, hierarchical layout modeling, and diffusion-based rendering. We evaluate both multi-stage and end-to-end OCR systems on MEH, studying the effects of synthetic pretraining, model scale, and cross-style transfer. To enable future expansion, we release pyThoth, an annotation tool with built-in model assistance for streamlined human-in-the-loop labeling. We believe that these contributions lay the groundwork for building an AI-powered research assistant for Egyptology.

4D Gaussian Splatting (4DGS) has recently emerged as a promising technique for capturing complex dynamic 3D scenes with high fidelity. It utilizes a 4D Gaussian representation and a GPU-friendly rasterizer, enabling rapid rendering speeds. Despite its advantages, 4DGS faces significant challenges, notably the requirement of millions of 4D Gaussians, each with extensive associated attributes, leading to substantial memory and storage cost. This paper introduces a memory-efficient framework for 4DGS. We streamline the color attribute by decomposing it into a per-Gaussian direct color component with only 3 parameters and a shared lightweight alternating current color predictor. This approach eliminates the need for spherical harmonics coefficients, which typically involve up to 144 parameters in classic 4DGS, thereby creating a memory-efficient 4D Gaussian representation. Furthermore, we introduce an entropy-constrained Gaussian deformation technique that uses a deformation field to expand the action range of each Gaussian and integrates an opacity-based entropy loss to limit the number of Gaussians, thus forcing our model to use as few Gaussians as possible to fit a dynamic scene well. With simple half-precision storage and zip compression, our framework achieves a storage reduction by approximately 190xand 125xon the Technicolor and Neural 3D Video datasets, respectively, compared to the original 4DGS. Meanwhile, it maintains comparable rendering speeds and scene representation quality, setting a new standard in the field. Code is available at https://github.com/Xinjie-Q/MEGA.

Video large language models (Video-LLMs) have made significant progress in understanding videos. However, processing multiple frames leads to lengthy visual token sequences, presenting challenges such as the limited context length cannot accommodate the entire video, and the inclusion of irrelevant frames hinders visual perception. Hence, effective frame selection is crucial. This paper emphasizes that frame selection should follow three key principles: query relevance, list-wise diversity, and sequentiality. Existing methods, such as uniform frame sampling and query-frame matching, do not capture all of these principles. Thus, we propose Markov decision determinantal point process with dynamic programming (MDP^3) for frame selection, a training-free and model-agnostic method that can be seamlessly integrated into existing Video-LLMs. Our method first estimates frame similarities conditioned on the query using a conditional Gaussian kernel within the reproducing kernel Hilbert space (RKHS). We then apply the determinantal point process (DPP) to the similarity matrix to capture both query relevance and list-wise diversity. To incorporate sequentiality, we segment the video and apply DPP within each segment, conditioned on the preceding segment selection, modeled as a Markov decision process (MDP) for allocating selection sizes across segments. Theoretically, MDP^3 provides a (1-1/e)-approximate solution to the NP-hard list-wise frame selection problem with pseudo-polynomial time complexity, demonstrating its efficiency. Empirically, MDP significantly outperforms existing methods, verifying its effectiveness and robustness.