arXiv subjects

Shin'ichi Satoh

Publications and source records attributed to Shin'ichi Satoh.

At least 19 recordsLinked to original sources

arXiv2026

Beyond Frame Selection: Rethinking Long-Video Understanding with MLLMs

Multimodal Large Language Models (MLLMs) have made strong progress in video understanding, yet long videos remain difficult: the visual token budget grows with video length, so temporally sparse evidence is easily lost. Existing methods compress the input through uniform sampling or frame selection, but these strategies optimize different objectives, either broad temporal coverage or local question relevance, and neither preserves both global storyline context and fine-grained evidence. We propose VideoRouter (VR), which rethinks long-video understanding as coordinating complementary evidence views rather than selecting a single subset of frames. VideoRouter first organizes each video into a question-agnostic temporal hierarchy that partitions it into coarse-to-fine temporally coherent segments. Upper-level nodes capture broad storyline context and event progression, while lower-level nodes preserve fine-grained local details and evidence-bearing moments. This gives rise to two complementary views: a global view for coverage-oriented reasoning and a local view for detail-oriented evidence recovery. We further introduce a verification-guided router that judges which view is better supported by its own selected evidence and decides the final answer. Across six backbones, routing improves over both views in all settings, and the choice of view is shown to be dataset-dependent, confirming that no single evidence granularity is universally preferable. On VideoMME, our method outperforms state-of-the-art frame selection methods by 2.5 points, under the LLaVA-Video-7B backbone. We will release the code.

cs.CV↗

arXiv2026

Temporal Tree of Thought: Reasoning-Guided Visual Cue Search for Long-Video Understanding

Long-video understanding remains challenging for Multimodal Large Language Models (MLLMs) due to limited context length. Uniform sampling may miss crucial moments, while agent-based frame video understanding methods often evaluate frames independently, overlooking the temporal organization of videos. Ideally, evidence selection should mimic how humans answer questions about long videos: first locating the relevant segment from the global context, then zooming into local objects and details. We propose Temporal Tree of Thought T^3, a training-free framework for adaptive coarse-to-fine long-video understanding. T^3 constructs a question-agnostic hierarchical temporal tree via recursive temporally constrained clustering, where each node represents a contiguous segment with an informative key frame. During inference, T^3 performs an answer-retrieve-explore loop: it reasons over coarse representative frames, generates a search statement when evidence is insufficient, and expands relevant branches for finer-grained evidence. This process adaptively shifts the search target from temporal regions to specific objects and visual details to help video understanding. Experiments on VideoMME, LongVideoBench, and LVBench show that T^3 improves Qwen2.5-VL-7B by 0.5%, 4.6%, and 4.4%, respectively, under the same frame budget, demonstrating the effectiveness of structured temporal reasoning.

cs.CV↗

arXiv2026

Reason, Reward, Refine: Step-Level Errors Corrections with Structured Feedback for Physics Reasoning in Small Language Models

Physics reasoning fails structurally in small language models: an error at any step propagates forward, corrupting every inference that follows. Limited domain knowledge, hallucination under multi-step derivation, and distributional sensitivity compound this failure. We propose a step-level reward framework that identifies the first reasoning error, generates targeted structured feedback, and trains the model to revise its solution via policy gradient with KL regularization, without exposing it to ground truth solutions as generation targets. Unlike annotation-dependent step-level methods, no preference data construction is required and the external verifier operates exclusively at training time. Across five physics benchmarks, our framework delivers accuracy gains of 17-20% over CoT prompting and 10-16% over the strongest baseline, reduces calculation errors from 56.9% to 23.5%, and reduces miscomprehension errors from 22.3% to 12.0% in the best observed cases. Conceptual errors reduce from 89.7% to 68.7%, yet persist as the hardest failure mode across all conditions.

cs.AI↗

arXiv2026

DiverXplorer: Stock Image Exploration via Diversity Adjustment for Graphic Design

Graphic designers explore large stock image collections during open-ended or early-stage design tasks, yet common tools emphasize relevance and similarity, limiting designers' ability to overview the design space or discover visual patterns. We present an image exploration prototype that enables stepwise adjustment of diversity, allowing users to transition from diverse overviews to increasingly focused subsets during exploration. Our approach implements diversity control via determinantal point process (DPP)-based sampling and exposes diversity-similarity tradeoffs through interaction rather than static ranking. We report findings from a pilot study with professional graphic designers comparing our technique to baselines inspired by current tools in open-ended image selection tasks. Results suggest that stepwise diversity control supports early-stage sensemaking and comparison of visual patterns, while revealing important tradeoffs: diversity aids discovery and reduces backtracking, but becomes less desirable as exploration progresses. We aim to provide a novel perspective on how to implement transitions between diversity and similarity. Our code is available at https://github.com/CyberAgentAILab/DiverXplorer.

cs.HC↗

arXiv2025

Uplifting Table Tennis: A Robust, Real-World Application for 3D Trajectory and Spin Estimation

Obtaining the precise 3D motion of a table tennis ball from standard monocular videos is a challenging problem, as existing methods trained on synthetic data struggle to generalize to the noisy, imperfect ball and table detections of the real world. This is primarily due to the inherent lack of 3D ground truth trajectories and spin annotations for real-world video. To overcome this, we propose a novel two-stage pipeline that divides the problem into a front-end perception task and a back-end 2D-to-3D uplifting task. This separation allows us to train the front-end components with abundant 2D supervision from our newly created TTHQ dataset, while the back-end uplifting network is trained exclusively on physically-correct synthetic data. We specifically re-engineer the uplifting model to be robust to common real-world artifacts, such as missing detections and varying frame rates. By integrating a ball detector and a table keypoint detector, our approach transforms a proof-of-concept uplifting method into a practical, robust, and high-performing end-to-end application for 3D table tennis trajectory and spin analysis.

cs.CV↗

arXiv2025

Fairness Without Labels: Pseudo-Balancing for Bias Mitigation in Face Gender Classification

Face gender classification models often reflect and amplify demographic biases present in their training data, leading to uneven performance across gender and racial subgroups. We introduce pseudo-balancing, a simple and effective strategy for mitigating such biases in semi-supervised learning. Our method enforces demographic balance during pseudo-label selection, using only unlabeled images from a race-balanced dataset without requiring access to ground-truth annotations. We evaluate pseudo-balancing under two conditions: (1) fine-tuning a biased gender classifier using unlabeled images from the FairFace dataset, and (2) stress-testing the method with intentionally imbalanced training data to simulate controlled bias scenarios. In both cases, models are evaluated on the All-Age-Faces (AAF) benchmark, which contains a predominantly East Asian population. Our results show that pseudo-balancing consistently improves fairness while preserving or enhancing accuracy. The method achieves 79.81% overall accuracy - a 6.53% improvement over the baseline - and reduces the gender accuracy gap by 44.17%. In the East Asian subgroup, where baseline disparities exceeded 49%, the gap is narrowed to just 5.01%. These findings suggest that even in the absence of label supervision, access to a demographically balanced or moderately skewed unlabeled dataset can serve as a powerful resource for debiasing existing computer vision models.

cs.CV↗

arXiv2025

A Culturally-diverse Multilingual Multimodal Video Benchmark & Model

Large multimodal models (LMMs) have recently gained attention due to their effectiveness to understand and generate descriptions of visual content. Most existing LMMs are in English language. While few recent works explore multilingual image LMMs, to the best of our knowledge, moving beyond the English language for cultural and linguistic inclusivity is yet to be investigated in the context of video LMMs. In pursuit of more inclusive video LMMs, we introduce a multilingual Video LMM benchmark, named ViMUL-Bench, to evaluate Video LMMs across 14 languages, including both low- and high-resource languages: English, Chinese, Spanish, French, German, Hindi, Arabic, Russian, Bengali, Urdu, Sinhala, Tamil, Swedish, and Japanese. Our ViMUL-Bench is designed to rigorously test video LMMs across 15 categories including eight culturally diverse categories, ranging from lifestyles and festivals to foods and rituals and from local landmarks to prominent cultural personalities. ViMUL-Bench comprises both open-ended (short and long-form) and multiple-choice questions spanning various video durations (short, medium, and long) with 8k samples that are manually verified by native language speakers. In addition, we also introduce a machine translated multilingual video training set comprising 1.2 million samples and develop a simple multilingual video LMM, named ViMUL, that is shown to provide a better tradeoff between high-and low-resource languages for video understanding. We hope our ViMUL-Bench and multilingual video LMM along with a large-scale multilingual video training set will help ease future research in developing cultural and linguistic inclusive multilingual video LMMs. Our proposed benchmark, video LMM and training data will be publicly released at https://mbzuai-oryx.github.io/ViMUL/.

cs.CL↗

arXiv2025

Probabilistic Online Event Downsampling

Event cameras capture scene changes asynchronously on a per-pixel basis, enabling extremely high temporal resolution. However, this advantage comes at the cost of high bandwidth, memory, and computational demands. To address this, prior work has explored event downsampling, but most approaches rely on fixed heuristics or threshold-based strategies, limiting their adaptability. Instead, we propose a probabilistic framework, POLED, that models event importance through an event-importance probability density function (ePDF), which can be arbitrarily defined and adapted to different applications. Our approach operates in a purely online setting, estimating event importance on-the-fly from raw event streams, enabling scene-specific adaptation. Additionally, we introduce zero-shot event downsampling, where downsampled events must remain usable for models trained on the original event stream, without task-specific adaptation. We design a contour-preserving ePDF that prioritizes structurally important events and evaluate our method across four datasets and tasks--object classification, image interpolation, surface normal estimation, and object detection--demonstrating that intelligent sampling is crucial for maintaining performance under event-budget constraints. Code available.

cs.CV↗

arXiv2025

Foundation Model Insights and a Multi-Model Approach for Superior Fine-Grained One-shot Subset Selection

One-shot subset selection serves as an effective tool to reduce deep learning training costs by identifying an informative data subset based on the information extracted by an information extractor (IE). Traditional IEs, typically pre-trained on the target dataset, are inherently dataset-dependent. Foundation models (FMs) offer a promising alternative, potentially mitigating this limitation. This work investigates two key questions: (1) Can FM-based subset selection outperform traditional IE-based methods across diverse datasets? (2) Do all FMs perform equally well as IEs for subset selection? Extensive experiments uncovered surprising insights: FMs consistently outperform traditional IEs on fine-grained datasets, whereas their advantage diminishes on coarse-grained datasets with noisy labels. Motivated by these finding, we propose RAM-APL (RAnking Mean-Accuracy of Pseudo-class Labels), a method tailored for fine-grained image datasets. RAM-APL leverages multiple FMs to enhance subset selection by exploiting their complementary strengths. Our approach achieves state-of-the-art performance on fine-grained datasets, including Oxford-IIIT Pet, Food-101, and Caltech-UCSD Birds-200-2011.

cs.CV↗

arXiv2025

ReSeDis: A Dataset for Referring-based Object Search across Large-Scale Image Collections

Large-scale visual search engines are expected to solve a dual problem at once: (i) locate every image that truly contains the object described by a sentence and (ii) identify the object's bounding box or exact pixels within each hit. Existing techniques address only one side of this challenge. Visual grounding yields tight boxes and masks but rests on the unrealistic assumption that the object is present in every test image, producing a flood of false alarms when applied to web-scale collections. Text-to-image retrieval excels at sifting through massive databases to rank relevant images, yet it stops at whole-image matches and offers no fine-grained localization. We introduce Referring Search and Discovery (ReSeDis), the first task that unifies corpus-level retrieval with pixel-level grounding. Given a free-form description, a ReSeDis model must decide whether the queried object appears in each image and, if so, where it is, returning bounding boxes or segmentation masks. To enable rigorous study, we curate a benchmark in which every description maps uniquely to object instances scattered across a large, diverse corpus, eliminating unintended matches. We further design a task-specific metric that jointly scores retrieval recall and localization precision. Finally, we provide a straightforward zero-shot baseline using a frozen vision-language model, revealing significant headroom for future study. ReSeDis offers a realistic, end-to-end testbed for building the next generation of robust and scalable multimodal search systems.

cs.CV↗

arXiv2025

TransRef: Multi-Scale Reference Embedding Transformer for Reference-Guided Image Inpainting

Image inpainting for completing complicated semantic environments and diverse hole patterns of corrupted images is challenging even for state-of-the-art learning-based inpainting methods trained on large-scale data. A reference image capturing the same scene of a corrupted image offers informative guidance for completing the corrupted image as it shares similar texture and structure priors to that of the holes of the corrupted image. In this work, we propose a transformer-based encoder-decoder network, named TransRef, for reference-guided image inpainting. Specifically, the guidance is conducted progressively through a reference embedding procedure, in which the referencing features are subsequently aligned and fused with the features of the corrupted image. For precise utilization of the reference features for guidance, a reference-patch alignment (Ref-PA) module is proposed to align the patch features of the reference and corrupted images and harmonize their style differences, while a reference-patch transformer (Ref-PT) module is proposed to refine the embedded reference feature. Moreover, to facilitate the research of reference-guided image restoration tasks, we construct a publicly accessible benchmark dataset containing 50K pairs of input and reference images. Both quantitative and qualitative evaluations demonstrate the efficacy of the reference information and the proposed method over the state-of-the-art methods in completing complex holes. Code and dataset can be accessed at https://github.com/Cameltr/TransRef.

cs.CV↗

arXiv2024

Physical Adversarial Attack meets Computer Vision: A Decade Survey

Despite the impressive achievements of Deep Neural Networks (DNNs) in computer vision, their vulnerability to adversarial attacks remains a critical concern. Extensive research has demonstrated that incorporating sophisticated perturbations into input images can lead to a catastrophic degradation in DNNs' performance. This perplexing phenomenon not only exists in the digital space but also in the physical world. Consequently, it becomes imperative to evaluate the security of DNNs-based systems to ensure their safe deployment in real-world scenarios, particularly in security-sensitive applications. To facilitate a profound understanding of this topic, this paper presents a comprehensive overview of physical adversarial attacks. Firstly, we distill four general steps for launching physical adversarial attacks. Building upon this foundation, we uncover the pervasive role of artifacts carrying adversarial perturbations in the physical world. These artifacts influence each step. To denote them, we introduce a new term: adversarial medium. Then, we take the first step to systematically evaluate the performance of physical adversarial attacks, taking the adversarial medium as a first attempt. Our proposed evaluation metric, hiPAA, comprises six perspectives: Effectiveness, Stealthiness, Robustness, Practicability, Aesthetics, and Economics. We also provide comparative results across task categories, together with insightful observations and suggestions for future research directions.

cs.CV↗

arXiv2024

Matting by Generation

This paper introduces an innovative approach for image matting that redefines the traditional regression-based task as a generative modeling challenge. Our method harnesses the capabilities of latent diffusion models, enriched with extensive pre-trained knowledge, to regularize the matting process. We present novel architectural innovations that empower our model to produce mattes with superior resolution and detail. The proposed method is versatile and can perform both guidance-free and guidance-based image matting, accommodating a variety of additional cues. Our comprehensive evaluation across three benchmark datasets demonstrates the superior performance of our approach, both quantitatively and qualitatively. The results not only reflect our method's robust effectiveness but also highlight its ability to generate visually compelling mattes that approach photorealistic quality. The project page for this paper is available at https://lightchaserx.github.io/matting-by-generation/

cs.CV↗

arXiv2024

Certified Zeroth-order Black-Box Defense with Robust UNet Denoiser

Certified defense methods against adversarial perturbations have been recently investigated in the black-box setting with a zeroth-order (ZO) perspective. However, these methods suffer from high model variance with low performance on high-dimensional datasets due to the ineffective design of the denoiser and are limited in their utilization of ZO techniques. To this end, we propose a certified ZO preprocessing technique for removing adversarial perturbations from the attacked image in the black-box setting using only model queries. We propose a robust UNet denoiser (RDUNet) that ensures the robustness of black-box models trained on high-dimensional datasets. We propose a novel black-box denoised smoothing (DS) defense mechanism, ZO-RUDS, by prepending our RDUNet to the black-box model, ensuring black-box defense. We further propose ZO-AE-RUDS in which RDUNet followed by autoencoder (AE) is prepended to the black-box model. We perform extensive experiments on four classification datasets, CIFAR-10, CIFAR-10, Tiny Imagenet, STL-10, and the MNIST dataset for image reconstruction tasks. Our proposed defense methods ZO-RUDS and ZO-AE-RUDS beat SOTA with a huge margin of $35\%$ and $9\%$, for low dimensional (CIFAR-10) and with a margin of $20.61\%$ and $23.51\%$ for high-dimensional (STL-10) datasets, respectively.

cs.CV↗

arXiv2024

The SkatingVerse Workshop & Challenge: Methods and Results

The SkatingVerse Workshop & Challenge aims to encourage research in developing novel and accurate methods for human action understanding. The SkatingVerse dataset used for the SkatingVerse Challenge has been publicly released. There are two subsets in the dataset, i.e., the training subset and testing subset. The training subsets consists of 19,993 RGB video sequences, and the testing subsets consists of 8,586 RGB video sequences. Around 10 participating teams from the globe competed in the SkatingVerse Challenge. In this paper, we provide a brief summary of the SkatingVerse Workshop & Challenge including brief introductions to the top three methods. The submission leaderboard will be reopened for researchers that are interested in the human action understanding challenge. The benchmark dataset and other information can be found at: https://skatingverse.github.io/.

cs.CV↗

arXiv2024

RanLayNet: A Dataset for Document Layout Detection used for Domain Adaptation and Generalization

Large ground-truth datasets and recent advances in deep learning techniques have been useful for layout detection. However, because of the restricted layout diversity of these datasets, training on them requires a sizable number of annotated instances, which is both expensive and time-consuming. As a result, differences between the source and target domains may significantly impact how well these models function. To solve this problem, domain adaptation approaches have been developed that use a small quantity of labeled data to adjust the model to the target domain. In this research, we introduced a synthetic document dataset called RanLayNet, enriched with automatically assigned labels denoting spatial positions, ranges, and types of layout elements. The primary aim of this endeavor is to develop a versatile dataset capable of training models with robustness and adaptability to diverse document formats. Through empirical experimentation, we demonstrate that a deep layout identification model trained on our dataset exhibits enhanced performance compared to a model trained solely on actual documents. Moreover, we conduct a comparative analysis by fine-tuning inference models using both PubLayNet and IIIT-AR-13K datasets on the Doclaynet dataset. Our findings emphasize that models enriched with our dataset are optimal for tasks such as achieving 0.398 and 0.588 mAP95 score in the scientific document domain for the TABLE class.

cs.CV↗

arXiv2024

TC-OCR: TableCraft OCR for Efficient Detection & Recognition of Table Structure & Content

The automatic recognition of tabular data in document images presents a significant challenge due to the diverse range of table styles and complex structures. Tables offer valuable content representation, enhancing the predictive capabilities of various systems such as search engines and Knowledge Graphs. Addressing the two main problems, namely table detection (TD) and table structure recognition (TSR), has traditionally been approached independently. In this research, we propose an end-to-end pipeline that integrates deep learning models, including DETR, CascadeTabNet, and PP OCR v2, to achieve comprehensive image-based table recognition. This integrated approach effectively handles diverse table styles, complex structures, and image distortions, resulting in improved accuracy and efficiency compared to existing methods like Table Transformers. Our system achieves simultaneous table detection (TD), table structure recognition (TSR), and table content recognition (TCR), preserving table structures and accurately extracting tabular data from document images. The integration of multiple models addresses the intricacies of table recognition, making our approach a promising solution for image-based table understanding, data extraction, and information retrieval applications. Our proposed approach achieves an IOU of 0.96 and an OCR Accuracy of 78%, showcasing a remarkable improvement of approximately 25% in the OCR Accuracy compared to the previous Table Transformer approach.

cs.CV↗

arXiv2024

The Effects of Short Video-Sharing Services on Video Copy Detection

The short video-sharing services that allow users to post 10-30 second videos (e.g., YouTube Shorts and TikTok) have attracted a lot of attention in recent years. However, conventional video copy detection (VCD) methods mainly focus on general video-sharing services (e.g., YouTube and Bilibili), and the effects of short video-sharing services on video copy detection are still unclear. Considering that illegally copied videos in short video-sharing services have service-distinctive characteristics, especially in those time lengths, the pros and cons of VCD in those services are required to be analyzed. In this paper, we examine the effects of short video-sharing services on VCD by constructing a dataset that has short video-sharing service characteristics. Our novel dataset is automatically constructed from the publicly available dataset to have reference videos and fixed short-time-length query videos, and such automation procedures assure the reproducibility and data privacy preservation of this paper. From the experimental results focusing on segment-level and video-level situations, we can see that three effects: "Segment-level VCD in short video-sharing services is more difficult than those in general video-sharing services", "Video-level VCD in short video-sharing services is easier than those in general video-sharing services", "The video alignment component mainly suppress the detection performance in short video-sharing services".

cs.CV↗