Kiyoharu Aizawa

Atsuyuki Miyai, Qing Yu, Go Irie et al.

We present a novel vision-language prompt learning approach for few-shot out-of-distribution (OOD) detection. Few-shot OOD detection aims to detect OOD images from classes that are unseen during training using only a few labeled in-distribution (ID) images. While prompt learning methods such as CoOp have shown effectiveness and efficiency in few-shot ID classification, they still face limitations in OOD detection due to the potential presence of ID-irrelevant information in text embeddings. To address this issue, we introduce a new approach called Local regularized Context Optimization (LoCoOp), which performs OOD regularization that utilizes the portions of CLIP local features as OOD features during training. CLIP's local features have a lot of ID-irrelevant nuisances (e.g., backgrounds), and by learning to push them away from the ID class text embeddings, we can remove the nuisances in the ID class text embeddings and enhance the separation between ID and OOD. Experiments on the large-scale ImageNet OOD detection benchmarks demonstrate the superiority of our LoCoOp over zero-shot, fully supervised detection methods and prompt learning methods. Notably, even in a one-shot setting -- just one label per class, LoCoOp outperforms existing zero-shot and fully supervised detection methods. The code will be available via https://github.com/AtsuMiyai/LoCoOp.

Atsuyuki Miyai, Jingkang Yang, Jingyang Zhang et al.

Detecting out-of-distribution (OOD) samples is crucial for ensuring the safety of machine learning systems and has shaped the field of OOD detection. Meanwhile, several other problems are closely related to OOD detection, including anomaly detection (AD), novelty detection (ND), open set recognition (OSR), and outlier detection (OD). To unify these problems, a generalized OOD detection framework was proposed, taxonomically categorizing these five problems. However, Vision Language Models (VLMs) such as CLIP have significantly changed the paradigm and blurred the boundaries between these fields, again confusing researchers. In this survey, we first present a generalized OOD detection v2, encapsulating the evolution of these fields in the VLM era. Our framework reveals that, with some field inactivity and integration, the demanding challenges have become OOD detection and AD. Then, we highlight the significant shift in the definition, problem settings, and benchmarks; we thus feature a comprehensive review of the methodology for OOD detection and related tasks to clarify their relationship to OOD detection. Finally, we explore the advancements in the emerging Large Vision Language Model (LVLM) era, such as GPT-4V. We conclude with open challenges and future directions. The resource is available at https://github.com/AtsuMiyai/Awesome-OOD-VLM.

Atsuyuki Miyai, Qing Yu, Go Irie et al.

Zero-shot out-of-distribution (OOD) detection is a task that detects OOD images during inference with only in-distribution (ID) class names. Existing methods assume ID images contain a single, centered object, and do not consider the more realistic multi-object scenarios, where both ID and OOD objects are present. To meet the needs of many users, the detection method must have the flexibility to adapt the type of ID images. To this end, we present Global-Local Maximum Concept Matching (GL-MCM), which incorporates local image scores as an auxiliary score to enhance the separability of global and local visual features. Due to the simple ensemble score function design, GL-MCM can control the type of ID images with a single weight parameter. Experiments on ImageNet and multi-object benchmarks demonstrate that GL-MCM outperforms baseline zero-shot methods and is comparable to fully supervised methods. Furthermore, GL-MCM offers strong flexibility in adjusting the target type of ID images. The code is available via https://github.com/AtsuMiyai/GL-MCM.

Daichi Horita, Jiaolong Yang, Dong Chen et al.

Image completion techniques have made significant progress in filling missing regions (i.e., holes) in images. However, large-hole completion remains challenging due to limited structural information. In this paper, we address this problem by integrating explicit structural guidance into diffusion-based image completion, forming our structure-guided diffusion model (SGDM). It consists of two cascaded diffusion probabilistic models: structure and texture generators. The structure generator generates an edge image representing plausible structures within the holes, which is then used for guiding the texture generation process. To train both generators jointly, we devise a novel strategy that leverages optimal Bayesian denoising, which denoises the output of the structure generator in a single step and thus allows backpropagation. Our diffusion-based approach enables a diversity of plausible completions, while the editable edges allow for editing parts of an image. Our experiments on natural scene (Places) and face (CelebA-HQ) datasets demonstrate that our method achieves a superior or comparable visual quality compared to state-of-the-art approaches. The code is available for research purposes at https://github.com/UdonDa/Structure_Guided_Diffusion_Model.

Koki Tsubota, Hiroaki Akutsu, Kiyoharu Aizawa

Deep image compression performs better than conventional codecs, such as JPEG, on natural images. However, deep image compression is learning-based and encounters a problem: the compression performance deteriorates significantly for out-of-domain images. In this study, we highlight this problem and address a novel task: universal deep image compression. This task aims to compress images belonging to arbitrary domains, such as natural images, line drawings, and comics. To address this problem, we propose a content-adaptive optimization framework; this framework uses a pre-trained compression model and adapts the model to a target image during compression. Adapters are inserted into the decoder of the model. For each input image, our framework optimizes the latent representation extracted by the encoder and the adapter parameters in terms of rate-distortion. The adapter parameters are additionally transmitted per image. For the experiments, a benchmark dataset containing uncompressed images of four domains (natural images, line drawings, comics, and vector arts) is constructed and the proposed universal deep compression is evaluated. Finally, the proposed model is compared with non-adaptive and existing adaptive compression models. The comparison reveals that the proposed model outperforms these. The code and dataset are publicly available at https://github.com/kktsubota/universal-dic.

Jeonghun Baek, Yusuke Matsui, Kiyoharu Aizawa

Recognizing irregular texts has been a challenging topic in text recognition. To encourage research on this topic, we provide a novel comic onomatopoeia dataset (COO), which consists of onomatopoeia texts in Japanese comics. COO has many arbitrary texts, such as extremely curved, partially shrunk texts, or arbitrarily placed texts. Furthermore, some texts are separated into several parts. Each part is a truncated text and is not meaningful by itself. These parts should be linked to represent the intended meaning. Thus, we propose a novel task that predicts the link between truncated texts. We conduct three tasks to detect the onomatopoeia region and capture its intended meaning: text detection, text recognition, and link prediction. Through extensive experiments, we analyze the characteristics of the COO. Our data and code are available at \url{https://github.com/ku21fan/COO-Comic-Onomatopoeia}.

Atsuyuki Miyai, Qing Yu, Daiki Ikami et al.

Rotation is frequently listed as a candidate for data augmentation in contrastive learning but seldom provides satisfactory improvements. We argue that this is because the rotated image is always treated as either positive or negative. The semantics of an image can be rotation-invariant or rotation-variant, so whether the rotated image is treated as positive or negative should be determined based on the content of the image. Therefore, we propose a novel augmentation strategy, adaptive Positive or Negative Data Augmentation (PNDA), in which an original and its rotated image are a positive pair if they are semantically close and a negative pair if they are semantically different. To achieve PNDA, we first determine whether rotation is positive or negative on an image-by-image basis in an unsupervised way. Then, we apply PNDA to contrastive learning frameworks. Our experiments showed that PNDA improves the performance of contrastive learning. The code is available at \url{ https://github.com/AtsuMiyai/rethinking_rotation}.

Atsuyuki Miyai, Qing Yu, Go Irie et al.

Out-of-distribution (OOD) detection is critical for safety-sensitive machine learning applications and has been extensively studied, yielding a plethora of methods developed in the literature. However, most studies for OOD detection did not use pre-trained models and trained a backbone from scratch. In recent years, transferring knowledge from large pre-trained models to downstream tasks by lightweight tuning has become mainstream for training in-distribution (ID) classifiers. To bridge the gap between the practice of OOD detection and current classifiers, the unique and crucial problem is that the samples whose information networks know often come as OOD input. We consider that such data may significantly affect the performance of large pre-trained networks because the discriminability of these OOD data depends on the pre-training algorithm. Here, we define such OOD data as PT-OOD (Pre-Trained OOD) data. In this paper, we aim to reveal the effect of PT-OOD on the OOD detection performance of pre-trained networks from the perspective of pre-training algorithms. To achieve this, we explore the PT-OOD detection performance of supervised and self-supervised pre-training algorithms with linear-probing tuning, the most common efficient tuning method. Through our experiments and analysis, we find that the low linear separability of PT-OOD in the feature space heavily degrades the PT-OOD detection performance, and self-supervised models are more vulnerable to PT-OOD than supervised pre-trained models, even with state-of-the-art detection methods. To solve this vulnerability, we further propose a unique solution to large-scale pre-trained models: Leveraging powerful instance-by-instance discriminative representations of pre-trained models and detecting OOD in the feature space independent of the ID decision boundaries. The code will be available via https://github.com/AtsuMiyai/PT-OOD.

Yingxuan Li, Kiyoharu Aizawa, Yusuke Matsui

The expanding market for e-comics has spurred interest in the development of automated methods to analyze comics. For further understanding of comics, an automated approach is needed to link text in comics to characters speaking the words. Comics speaker detection research has practical applications, such as automatic character assignment for audiobooks, automatic translation according to characters' personalities, and inference of character relationships and stories. To deal with the problem of insufficient speaker-to-text annotations, we created a new annotation dataset Manga109Dialog based on Manga109. Manga109Dialog is the world's largest comics speaker annotation dataset, containing 132,692 speaker-to-text pairs. We further divided our dataset into different levels by prediction difficulties to evaluate speaker detection methods more appropriately. Unlike existing methods mainly based on distances, we propose a deep learning-based method using scene graph generation models. Due to the unique features of comics, we enhance the performance of our proposed model by considering the frame reading order. We conducted experiments using Manga109Dialog and other datasets. Experimental results demonstrate that our scene-graph-based approach outperforms existing methods, achieving a prediction accuracy of over 75%.

Qing Yu, Go Irie, Kiyoharu Aizawa

Unsupervised domain adaptation (UDA) has proven to be very effective in transferring knowledge obtained from a source domain with labeled data to a target domain with unlabeled data. Owing to the lack of labeled data in the target domain and the possible presence of unknown classes, open-set domain adaptation (ODA) has emerged as a potential solution to identify these classes during the training phase. Although existing ODA approaches aim to solve the distribution shifts between the source and target domains, most methods fine-tuned ImageNet pre-trained models on the source domain with the adaptation on the target domain. Recent visual-language foundation models (VLFM), such as Contrastive Language-Image Pre-Training (CLIP), are robust to many distribution shifts and, therefore, should substantially improve the performance of ODA. In this work, we explore generic ways to adopt CLIP, a popular VLFM, for ODA. We investigate the performance of zero-shot prediction using CLIP, and then propose an entropy optimization strategy to assist the ODA models with the outputs of CLIP. The proposed approach achieves state-of-the-art results on various benchmarks, demonstrating its effectiveness in addressing the ODA problem.

Haruka Aoki, Kiyoharu Aizawa

Designing fonts for Chinese characters is highly labor-intensive and time-consuming. While the latest methods successfully generate the English alphabet vector font, despite the high demand for automatic font generation, Chinese vector font generation has been an unsolved problem owing to its complex shape and numerous characters. This study addressed the problem of automatically generating Chinese vector fonts from only a single style and content reference. We proposed a novel network architecture with Transformer and loss functions to capture structural features without differentiable rendering. Although the dataset range was still limited to the sans-serif family, we successfully generated the Chinese vector font for the first time using the proposed method.

Yuuki Sawabe, Satoshi Ikehata, Kiyoharu Aizawa

360° images are informative -- it contains omnidirectional visual information around the camera. However, the areas that cover a 360° image is much larger than the human's field of view, therefore important information in different view directions is easily overlooked. To tackle this issue, we propose a method for predicting the optimal set of Region of Interest (RoI) from a single 360° image using the visual saliency as a clue. To deal with the scarce, strongly biased training data of existing single 360° image saliency prediction dataset, we also propose a data augmentation method based on the spherical random data rotation. From the predicted saliency map and redundant candidate regions, we obtain the optimal set of RoIs considering both the saliency within a region and the Interaction-Over-Union (IoU) between regions. We conduct the subjective evaluation to show that the proposed method can select regions that properly summarize the input 360° image.

Takashi Otonari, Satoshi Ikehata, Kiyoharu Aizawa

In recent years, the performance of novel view synthesis using perspective images has dramatically improved with the advent of neural radiance fields (NeRF). This study proposes two novel techniques that effectively build NeRF for 360{\textdegree} omnidirectional images. Due to the characteristics of a 360{\textdegree} image of ERP format that has spatial distortion in their high latitude regions and a 360{\textdegree} wide viewing angle, NeRF's general ray sampling strategy is ineffective. Hence, the view synthesis accuracy of NeRF is limited and learning is not efficient. We propose two non-uniform ray sampling schemes for NeRF to suit 360{\textdegree} images - distortion-aware ray sampling and content-aware ray sampling. We created an evaluation dataset Synth360 using Replica and SceneCity models of indoor and outdoor scenes, respectively. In experiments, we show that our proposal successfully builds 360{\textdegree} image NeRF in terms of both accuracy and efficiency. The proposal is widely applicable to advanced variants of NeRF. DietNeRF, AugNeRF, and NeRF++ combined with the proposed techniques further improve the performance. Moreover, we show that our proposed method enhances the quality of real-world scenes in 360{\textdegree} images. Synth360: https://drive.google.com/drive/folders/1suL9B7DO2no21ggiIHkH3JF3OecasQLb.

Yuya Hasegawa, Ikehata Satoshi, Kiyoharu Aizawa

360° images are widely available over the last few years. This paper proposes a new technique for single 360° image depth prediction under open environments. Depth prediction from a 360° single image is not easy for two reasons. One is the limitation of supervision datasets - the currently available dataset is limited to indoor scenes. The other is the problems caused by Equirectangular Projection Format (ERP), commonly used for 360° images, that are coordinate and distortion. There is only one method existing that uses cube map projection to produce six perspective images and apply self-supervised learning using motion pictures for perspective depth prediction to deal with these problems. Different from the existing method, we directly use the ERP format. We propose a framework of direct use of ERP with coordinate conversion of correspondences and distortion-aware upsampling module to deal with the ERP related problems and extend a self-supervised learning method for open environments. For the experiments, we firstly built a dataset for the evaluation, and quantitatively evaluate the depth prediction in outdoor scenes. We show that it outperforms the state-of-the-art technique

Atsuyuki Miyai, Shota Onohara, Jeonghun Baek et al.

This paper introduces JMMMU-Pro, an image-based Japanese Multi-discipline Multimodal Understanding Benchmark, and Vibe Benchmark Construction, a scalable construction method. Following the evolution from MMMU to MMMU-Pro, JMMMU-Pro extends JMMMU by composing the question image and question text into a single image, thereby creating a benchmark that requires integrated visual-textual understanding through visual perception. To build JMMMU-Pro, we propose Vibe Benchmark Construction, a methodology in which an image generative model (e.g., Nano Banana Pro) produces candidate visual questions, and humans verify the outputs and, when necessary, regenerate with adjusted prompts to ensure quality. By leveraging Nano Banana Pro's highly realistic image generation capabilities and its ability to embed clean Japanese text, we construct a high-quality benchmark at low cost, covering a wide range of background and layout designs. Experimental results show that all open-source LMMs struggle substantially with JMMMU-Pro, underscoring JMMMU-Pro as an important benchmark for guiding future efforts in the open-source community. We believe that JMMMU-Pro provides a more rigorous evaluation tool for assessing the Japanese capabilities of LMMs and that our Vibe Benchmark Construction also offers an efficient guideline for future development of image-based VQA benchmarks.

Mitsuki Watanabe, Sosuke Amano, Kiyoharu Aizawa et al.

Food image classification models are crucial for dietary management applications because they reduce the burden of manual meal logging. However, most publicly available datasets for training such models rely on web-crawled images, which often differ from users' real-world meal photos. In this work, we present FoodLogAthl-218, a food image dataset constructed from real-world meal records collected through the dietary management application FoodLog Athl. The dataset contains 6,925 images across 218 food categories, with a total of 14,349 bounding boxes. Rich metadata, including meal date and time, anonymized user IDs, and meal-level context, accompany each image. Unlike conventional datasets-where a predefined class set guides web-based image collection-our data begins with user-submitted photos, and labels are applied afterward. This yields greater intra-class diversity, a natural frequency distribution of meal types, and casual, unfiltered images intended for personal use rather than public sharing. In addition to (1) a standard classification benchmark, we introduce two FoodLog-specific tasks: (2) an incremental fine-tuning protocol that follows the temporal stream of users' logs, and (3) a context-aware classification task where each image contains multiple dishes, and the model must classify each dish by leveraging the overall meal context. We evaluate these tasks using large multimodal models (LMMs). The dataset is publicly available at https://huggingface.co/datasets/FoodLog/FoodLogAthl-218.

Daichi Horita, Naoto Inoue, Kotaro Kikuchi et al.

Content-aware graphic layout generation aims to automatically arrange visual elements along with a given content, such as an e-commerce product image. In this paper, we argue that the current layout generation approaches suffer from the limited training data for the high-dimensional layout structure. We show that a simple retrieval augmentation can significantly improve the generation quality. Our model, which is named Retrieval-Augmented Layout Transformer (RALF), retrieves nearest neighbor layout examples based on an input image and feeds these results into an autoregressive generator. Our model can apply retrieval augmentation to various controllable generation tasks and yield high-quality layouts within a unified architecture. Our extensive experiments show that RALF successfully generates content-aware layouts in both constrained and unconstrained settings and significantly outperforms the baselines.

Koki Tsubota, Hiroaki Akutsu, Kiyoharu Aizawa

Image quality assessment (IQA) is a fundamental metric for image processing tasks (e.g., compression). With full-reference IQAs, traditional IQAs, such as PSNR and SSIM, have been used. Recently, IQAs based on deep neural networks (deep IQAs), such as LPIPS and DISTS, have also been used. It is known that image scaling is inconsistent among deep IQAs, as some perform down-scaling as pre-processing, whereas others instead use the original image size. In this paper, we show that the image scale is an influential factor that affects deep IQA performance. We comprehensively evaluate four deep IQAs on the same five datasets, and the experimental results show that image scale significantly influences IQA performance. We found that the most appropriate image scale is often neither the default nor the original size, and the choice differs depending on the methods and datasets used. We visualized the stability and found that PieAPP is the most stable among the four deep IQAs.

Naoki Sugimoto, Satoshi Ikehata, Kiyoharu Aizawa

Movie-Map, an interactive first-person-view map that engages the user in a simulated walking experience, comprises short 360° video segments separated by traffic intersections that are seamlessly connected according to the viewer's direction of travel. However, in wide urban-scale areas with numerous intersecting roads, manual intersection segmentation requires significant human effort. Therefore, automatic identification of intersections from 360° videos is an important problem for scaling up Movie-Map. In this paper, we propose a novel method that identifies an intersection from individual frames in 360° videos. Instead of formulating the intersection identification as a standard binary classification task with a 360° image as input, we identify an intersection based on the number of the possible directions of travel (PDoT) in perspective images projected in eight directions from a single 360° image detected by the neural network for handling various types of intersections. We constructed a large-scale 360° Image Intersection Identification (iii360) dataset for training and evaluation where 360° videos were collected from various areas such as school campus, downtown, suburb, and china town and demonstrate that our PDoT-based method achieves 88\% accuracy, which is significantly better than that achieved by the direct naive binary classification based method. The source codes and a partial dataset will be shared in the community after the paper is published.

Yuki Imajuku, Yoko Yamakata, Kiyoharu Aizawa

Research on food image understanding using recipe data has been a long-standing focus due to the diversity and complexity of the data. Moreover, food is inextricably linked to people's lives, making it a vital research area for practical applications such as dietary management. Recent advancements in Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities, not only in their vast knowledge but also in their ability to handle languages naturally. While English is predominantly used, they can also support multiple languages including Japanese. This suggests that MLLMs are expected to significantly improve performance in food image understanding tasks. We fine-tuned open MLLMs LLaVA-1.5 and Phi-3 Vision on a Japanese recipe dataset and benchmarked their performance against the closed model GPT-4o. We then evaluated the content of generated recipes, including ingredients and cooking procedures, using 5,000 evaluation samples that comprehensively cover Japanese food culture. Our evaluation demonstrates that the open models trained on recipe data outperform GPT-4o, the current state-of-the-art model, in ingredient generation. Our model achieved F1 score of 0.531, surpassing GPT-4o's F1 score of 0.481, indicating a higher level of accuracy. Furthermore, our model exhibited comparable performance to GPT-4o in generating cooking procedure text.

Sandra Zhang Ding, Jiafeng Mao, Kiyoharu Aizawa

Based on recent advanced diffusion models, Text-to-image (T2I) generation models have demonstrated their capabilities to generate diverse and high-quality images. However, leveraging their potential for real-world content creation, particularly in providing users with precise control over the image generation result, poses a significant challenge. In this paper, we propose an innovative training-free pipeline that extends existing text-to-image generation models to incorporate a sketch as an additional condition. To generate new images with a layout and structure closely resembling the input sketch, we find that these core features of a sketch can be tracked with the cross-attention maps of diffusion models. We introduce latent optimization, a method that refines the noisy latent at each intermediate step of the generation process using cross-attention maps to ensure that the generated images adhere closely to the desired structure outlined in the reference sketch. Through latent optimization, our method enhances the accuracy of image generation, offering users greater control and customization options in content creation.

Jeonghun Baek, Atsuyuki Miyai, Shota Onohara et al.

Manga is a culturally distinctive multimodal medium and one of the most influential forms of Japanese popular culture. As AI systems increasingly target manga understanding, OCR, and translation, Manga109 has become a foundational dataset for manga-related AI research. However, the current Manga109 dataset contains transcription errors and coarse annotations, which do not align well with modern OCR and multimodal manga understanding tasks. In this work, we revisit the dialogue text annotations of Manga109 and identify five categories of annotation issues, including transcription errors, missing text regions, overlapping dialogue and onomatopoeia, and under-segmented speech balloons. To address these issues, we combine OCR-based issue detection and manual revision to construct Manga109-v2026, revising approximately 29,000 dialogue annotations. Our revisions better align Manga109 with modern OCR and multimodal manga understanding systems while preserving expressive structures characteristic of manga.

Atsuyuki Miyai, Mashiro Toyooka, Takashi Otonari et al.

Understanding the current capabilities and risks of AI Scientist systems is essential for ensuring trustworthy and sustainable AI-driven scientific progress while preserving the integrity of the academic ecosystem. To this end, we develop Jr. AI Scientist, a state-of-the-art autonomous AI scientist system that mimics the core research workflow of a novice student researcher: Given the baseline paper from the human mentor, it analyzes its limitations, formulates novel hypotheses for improvement, and iteratively conducts experiments until improvements are realized, and writes a paper with the results. Unlike previous approaches that assume full automation or operate on small-scale code, Jr. AI Scientist follows a well-defined research workflow and leverages modern coding agents to handle complex, multi-file implementations, leading to scientifically valuable contributions. Through our experiments, the Jr. AI Scientist successfully generated new research papers that build upon real NeurIPS, IJCV, and ICLR works by proposing and implementing novel methods. For evaluation, we conducted automated assessments using AI Reviewers, author-led evaluations, and submissions to Agents4Science, a venue dedicated to AI-driven scientific contributions. The findings demonstrate that Jr. AI Scientist generates papers receiving higher review scores than existing fully automated systems. Nevertheless, we identify important limitations from both the author evaluation and the Agents4Science reviews, indicating the potential risks of directly applying current AI Scientist systems and key challenges for future research. Finally, we comprehensively report various risks identified during development. We believe this study clarifies the current role and limitations of AI Scientist systems, offering insights into the areas that still require human expertise and the risks that may emerge as these systems evolve.

Hikaru Ikuta, Leslie Wöhler, Kiyoharu Aizawa

Manga is a popular medium that combines stylized drawings and text to convey stories. As manga panels differ from natural images, computational systems traditionally had to be designed specifically for manga. Recently, the adaptive nature of modern large multimodal models (LMMs) shows possibilities for more general approaches. To provide an analysis of the current capability of LMMs for manga understanding tasks and identifying areas for their improvement, we design and evaluate MangaUB, a novel manga understanding benchmark for LMMs. MangaUB is designed to assess the recognition and understanding of content shown in a single panel as well as conveyed across multiple panels, allowing for a fine-grained analysis of a model's various capabilities required for manga understanding. Our results show strong performance on the recognition of image content, while understanding the emotion and information conveyed across multiple panels is still challenging, highlighting future work towards LMMs for manga understanding.

Shiho Noda, Atsuyuki Miyai, Qing Yu et al.

Out-of-distribution (OOD) detection is a task that detects OOD samples during inference to ensure the safety of deployed models. However, conventional benchmarks have reached performance saturation, making it difficult to compare recent OOD detection methods. To address this challenge, we introduce three novel OOD detection benchmarks that enable a deeper understanding of method characteristics and reflect real-world conditions. First, we present ImageNet-X, designed to evaluate performance under challenging semantic shifts. Second, we propose ImageNet-FS-X for full-spectrum OOD detection, assessing robustness to covariate shifts (feature distribution shifts). Finally, we propose Wilds-FS-X, which extends these evaluations to real-world datasets, offering a more comprehensive testbed. Our experiments reveal that recent CLIP-based OOD detection methods struggle to varying degrees across the three proposed benchmarks, and none of them consistently outperforms the others. We hope the community goes beyond specific benchmarks and includes more challenging conditions reflecting real-world scenarios. The code is https://github.com/hoshi23/OOD-X-Benchmarks.

Atsuyuki Miyai, Jingkang Yang, Jingyang Zhang et al.

This paper introduces a novel task to evaluate the robust understanding capability of Large Multimodal Models (LMMs), termed $\textbf{Unsolvable Problem Detection (UPD)}$. Multiple-choice question answering (MCQA) is widely used to assess the understanding capability of LMMs, but it does not guarantee that LMMs truly comprehend the answer. UPD assesses the LMM's ability to withhold answers when encountering unsolvable problems of MCQA, verifying whether the model truly understands the answer. UPD encompasses three problems: Absent Answer Detection (AAD), Incompatible Answer Set Detection (IASD), and Incompatible Visual Question Detection (IVQD), covering unsolvable cases like answer-lacking or incompatible choices and image-question mismatches. For the evaluation, we introduce the MM-UPD Bench, a benchmark for assessing performance across various ability dimensions. Our experiments reveal that even most LMMs, which demonstrate adequate performance on existing benchmarks, struggle significantly with MM-UPD, underscoring a novel aspect of trustworthiness that current benchmarks have overlooked. A detailed analysis shows that LMMs have different bottlenecks and chain-of-thought and self-reflection improved performance for LMMs with the bottleneck in their LLM capability. We hope our insights will enhance the broader understanding and development of more reliable LMMs. The code is available at https://github.com/AtsuMiyai/UPD.

Jeonghun Baek, Yusuke Matsui, Kiyoharu Aizawa

In this paper, we investigate cross-lingual learning (CLL) for multilingual scene text recognition (STR). CLL transfers knowledge from one language to another. We aim to find the condition that exploits knowledge from high-resource languages for improving performance in low-resource languages. To do so, we first examine if two general insights about CLL discussed in previous works are applied to multilingual STR: (1) Joint learning with high- and low-resource languages may reduce performance on low-resource languages, and (2) CLL works best between typologically similar languages. Through extensive experiments, we show that two general insights may not be applied to multilingual STR. After that, we show that the crucial condition for CLL is the dataset size of high-resource languages regardless of the kind of high-resource languages. Our code, data, and models are available at https://github.com/ku21fan/CLL-STR.

Mashiro Toyooka, Kiyoharu Aizawa, Yoko Yamakata

Large Language Models (LLMs) are trained on a vast amount of procedural texts, but they do not directly observe real-world phenomena. In the context of cooking recipes, this poses a challenge, as intermediate states of ingredients are often omitted, making it difficult for models to track ingredient states and understand recipes accurately. In this paper, we apply state probing, a method for evaluating a language model's understanding of the world, to the domain of cooking. We propose a new task and dataset for evaluating how well LLMs can recognize intermediate ingredient states during cooking procedures. We first construct a new Japanese recipe dataset with clear and accurate annotations of ingredient state changes, collected from well-structured and controlled recipe texts. Using this dataset, we design three novel tasks to evaluate whether LLMs can track ingredient state transitions and identify ingredients present at intermediate steps. Our experiments with widely used LLMs, such as Llama3.1-70B and Qwen2.5-72B, show that learning ingredient state knowledge improves their understanding of cooking processes, achieving performance comparable to commercial LLMs. The dataset are publicly available at: https://huggingface.co/datasets/mashi6n/nhkrecipe-100-anno-1

Jeonghun Baek, Yusuke Matsui, Kiyoharu Aizawa

Scene text recognition (STR) task has a common practice: All state-of-the-art STR models are trained on large synthetic data. In contrast to this practice, training STR models only on fewer real labels (STR with fewer labels) is important when we have to train STR models without synthetic data: for handwritten or artistic texts that are difficult to generate synthetically and for languages other than English for which we do not always have synthetic data. However, there has been implicit common knowledge that training STR models on real data is nearly impossible because real data is insufficient. We consider that this common knowledge has obstructed the study of STR with fewer labels. In this work, we would like to reactivate STR with fewer labels by disproving the common knowledge. We consolidate recently accumulated public real data and show that we can train STR models satisfactorily only with real labeled data. Subsequently, we find simple data augmentation to fully exploit real data. Furthermore, we improve the models by collecting unlabeled data and introducing semi- and self-supervised methods. As a result, we obtain a competitive model to state-of-the-art methods. To the best of our knowledge, this is the first study that 1) shows sufficient performance by only using real labels and 2) introduces semi- and self-supervised methods into STR with fewer labels. Our code and data are available: https://github.com/ku21fan/STR-Fewer-Labels

Junjun Jiang, Yi Yu, Suhua Tang et al.

Face hallucination is a technique that reconstruct high-resolution (HR) faces from low-resolution (LR) faces, by using the prior knowledge learned from HR/LR face pairs. Most state-of-the-arts leverage position-patch prior knowledge of human face to estimate the optimal representation coefficients for each image patch. However, they focus only the position information and usually ignore the context information of image patch. In addition, when they are confronted with misalignment or the Small Sample Size (SSS) problem, the hallucination performance is very poor. To this end, this study incorporates the contextual information of image patch and proposes a powerful and efficient context-patch based face hallucination approach, namely Thresholding Locality-constrained Representation and Reproducing learning (TLcR-RL). Under the context-patch based framework, we advance a thresholding based representation method to enhance the reconstruction accuracy and reduce the computational complexity. To further improve the performance of the proposed algorithm, we propose a promotion strategy called reproducing learning. By adding the estimated HR face to the training set, which can simulates the case that the HR version of the input LR face is present in the training set, thus iteratively enhancing the final hallucination result. Experiments demonstrate that the proposed TLcR-RL method achieves a substantial increase in the hallucinated results, both subjectively and objectively. Additionally, the proposed framework is more robust to face misalignment and the SSS problem, and its hallucinated HR face is still very good when the LR test face is from the real-world. The MATLAB source code is available at https://github.com/junjun-jiang/TLcR-RL

Akito Takeki, Daiki Ikami, Go Irie et al.

Convolutional neural network (CNN) architectures utilize downsampling layers, which restrict the subsequent layers to learn spatially invariant features while reducing computational costs. However, such a downsampling operation makes it impossible to use the full spectrum of input features. Motivated by this observation, we propose a novel layer called parallel grid pooling (PGP) which is applicable to various CNN models. PGP performs downsampling without discarding any intermediate feature. It works as data augmentation and is complementary to commonly used data augmentation techniques. Furthermore, we demonstrate that a dilated convolution can naturally be represented using PGP operations, which suggests that the dilated convolution can also be regarded as a type of data augmentation technique. Experimental results based on popular image classification benchmarks demonstrate the effectiveness of the proposed method. Code is available at: https://github.com/akitotakeki

Shota Onohara, Atsuyuki Miyai, Yuki Imajuku et al.

Accelerating research on Large Multimodal Models (LMMs) in non-English languages is crucial for enhancing user experiences across broader populations. In this paper, we introduce JMMMU (Japanese MMMU), the first large-scale Japanese benchmark designed to evaluate LMMs on expert-level tasks based on the Japanese cultural context. To facilitate comprehensive culture-aware evaluation, JMMMU features two complementary subsets: (i) culture-agnostic (CA) subset, where the culture-independent subjects (e.g., Math) are selected and translated into Japanese, enabling one-to-one comparison with its English counterpart MMMU; and (ii) culture-specific (CS) subset, comprising newly crafted subjects that reflect Japanese cultural context. Using the CA subset, we observe performance drop in many LMMs when evaluated in Japanese, which is purely attributable to language variation. Using the CS subset, we reveal their inadequate Japanese cultural understanding. Further, by combining both subsets, we identify that some LMMs perform well on the CA subset but not on the CS subset, exposing a shallow understanding of the Japanese language that lacks depth in cultural understanding. We hope this work will not only help advance LMM performance in Japanese but also serve as a guideline to create high-standard, culturally diverse benchmarks for multilingual LMM development. The project page is https://mmmu-japanese-benchmark.github.io/JMMMU/.

Yingxuan Li, Ryota Hinami, Kiyoharu Aizawa et al.

Recognizing characters and predicting speakers of dialogue are critical for comic processing tasks, such as voice generation or translation. However, because characters vary by comic title, supervised learning approaches like training character classifiers which require specific annotations for each comic title are infeasible. This motivates us to propose a novel zero-shot approach, allowing machines to identify characters and predict speaker names based solely on unannotated comic images. In spite of their importance in real-world applications, these task have largely remained unexplored due to challenges in story comprehension and multimodal integration. Recent large language models (LLMs) have shown great capability for text understanding and reasoning, while their application to multimodal content analysis is still an open problem. To address this problem, we propose an iterative multimodal framework, the first to employ multimodal information for both character identification and speaker prediction tasks. Our experiments demonstrate the effectiveness of the proposed framework, establishing a robust baseline for these tasks. Furthermore, since our method requires no training data or annotations, it can be used as-is on any comic series.

Jiafeng Mao, Xueting Wang, Kiyoharu Aizawa

Text-to-image diffusion models allow users control over the content of generated images. Still, text-to-image generation occasionally leads to generation failure requiring users to generate dozens of images under the same text prompt before they obtain a satisfying result. We formulate the lottery ticket hypothesis in denoising: randomly initialized Gaussian noise images contain special pixel blocks (winning tickets) that naturally tend to be denoised into specific content independently. The generation failure in standard text-to-image synthesis is caused by the gap between optimal and actual spatial distribution of winning tickets in initial noisy images. To this end, we implement semantic-driven initial image construction creating initial noise from known winning tickets for each concept mentioned in the prompt. We conduct a series of experiments that verify the properties of winning tickets and demonstrate their generalizability across images and prompts. Our results show that aggregating winning tickets into the initial noise image effectively induce the model to generate the specified object at the corresponding location. Project Page: https://ut-mao.github.io/noise.github.io

Atsuyuki Miyai, Zaiying Zhao, Kazuki Egashira et al.

Powered by a large language model (LLM), a web browsing agent operates web browsers in a human-like manner and offers a highly transparent path toward automating a wide range of everyday tasks. As web agents become increasingly capable and demonstrate proficiency in general browsing tasks, a critical question emerges: Can they go beyond general browsing to robustly handle tasks that are tedious and complex, or chores that humans often avoid doing themselves? In this paper, we introduce WebChoreArena, a new fully reproducible benchmark comprising 532 carefully curated tasks designed to extend the scope of WebArena beyond general browsing to more labor-intensive and tedious tasks. WebChoreArena systematically integrates three key challenges: (i) Massive Memory tasks requiring accurate retrieval of large amounts of information in the observations, (ii) Calculation tasks demanding precise mathematical reasoning, and (iii) Long-Term Memory tasks necessitating long-term memory across multiple webpages. Built on top of the fully reproducible and widely adopted four WebArena simulation environments, WebChoreArena ensures strict reproducibility and enables fair, direct comparisons with the established WebArena benchmark, offering key insights into agent progress. Our experimental results demonstrate that as LLMs evolve, represented by GPT-4o, Claude 3.7 Sonnet, and Gemini 2.5 Pro, significant improvements in performance are observed on WebChoreArena. These findings suggest that WebChoreArena is well-suited to measure the advancement of state-of-the-art LLMs with greater clarity. Nevertheless, the results also indicate that even with Gemini 2.5 Pro, there remains substantial room for improvement compared to WebArena, highlighting the increased challenges posed by WebChoreArena.

Takashi Otonari, Satoshi Ikehata, Kiyoharu Aizawa

Recent advancements in the study of Neural Radiance Fields (NeRF) for dynamic scenes often involve explicit modeling of scene dynamics. However, this approach faces challenges in modeling scene dynamics in urban environments, where moving objects of various categories and scales are present. In such settings, it becomes crucial to effectively eliminate moving objects to accurately reconstruct static backgrounds. Our research introduces an innovative method, termed here as Entity-NeRF, which combines the strengths of knowledge-based and statistical strategies. This approach utilizes entity-wise statistics, leveraging entity segmentation and stationary entity classification through thing/stuff segmentation. To assess our methodology, we created an urban scene dataset masked with moving objects. Our comprehensive experiments demonstrate that Entity-NeRF notably outperforms existing techniques in removing moving objects and reconstructing static urban backgrounds, both quantitatively and qualitatively.

Tatsuro Banno, Koki Kawada, Mizuki Takenawa et al.

Virtual flood experience systems, which enable users to vividly experience flooding, are attracting increasing attention as effective tools for communicating flood risks. However, existing systems typically rely on virtual cities that do not correspond to real locations and often lack sufficient photorealism, limiting users' ability to relate scenarios to their own surroundings. Although 360° video-based virtual environments offer a simple and scalable way to visually replicate real-world scenes, effective 3D flood visualization in these environments typically requires 3D building geometry of the target area, which is not readily available in many regions. To address this limitation, we propose a new virtual flood experience framework that integrates 360° videos with 3D models automatically constructed from widely available 2D building footprints. By extruding footprints to plausible heights and spatially aligning the constructed models with 360° videos, our framework enables 3D flood visualization in photorealistic environments without relying on pre-existing city models such as CityGML. We demonstrate the framework in Memuro, Hokkaido, Japan, an area vulnerable to river flooding. A user study with local residents showed that the proposed system enhances users' ability to envision location-specific flood evacuation situations, demonstrating its potential as an effective tool for disaster risk communication and education.

Tatsuki Kawakami, Kazuki Egashira, Atsuyuki Miyai et al.

In recent years, unlearning techniques, which are methods for inducing a model to "forget" previously learned information, have attracted attention as a way to address privacy and copyright concerns in large language models (LLMs) and large multimodal models (LMMs). While several unlearning benchmarks have been established for LLMs, a practical evaluation framework for unlearning in LMMs has been less explored. Specifically, existing unlearning benchmark for LMMs considers only scenarios in which the model is required to unlearn fine-tuned knowledge through a single unlearning operation. In this study, we introduce PULSE protocol for realistic unlearning scenarios for LMMs by introducing two critical perspectives: (i) Pre-trained knowledge Unlearning for analyzing the effect across different knowledge acquisition phases and (ii) Long-term Sustainability Evaluation to address sequential requests. We then evaluate existing unlearning methods along these dimensions. Our results reveal that, although some techniques can successfully unlearn knowledge acquired through fine-tuning, they struggle to eliminate information learned during pre-training. Moreover, methods that effectively unlearn a batch of target data in a single operation exhibit substantial performance degradation when the same data are split and unlearned sequentially.

Atsuyuki Miyai, Mashiro Toyooka, Zaiying Zhao et al.

This paper introduces the first systematic evaluation framework for quantifying the quality and risks of papers written by modern coding agents. While AI-driven paper writing has become a growing concern, rigorous evaluation of the quality and potential risks of AI-written papers remains limited, and a unified understanding of their reliability is still lacking. We introduce Paper Reconstruction Evaluation (PaperRecon), an evaluation framework in which an overview (overview.md) is created from an existing paper, after which an agent generates a full paper based on the overview and minimal additional resources, and the result is subsequently compared against the original paper. PaperRecon disentangles the evaluation of the AI-written papers into two orthogonal dimensions, Presentation and Hallucination, where Presentation is evaluated using a rubric and Hallucination is assessed via agentic evaluation grounded in the original paper source. For evaluation, we introduce PaperWrite-Bench, a benchmark of 51 papers from top-tier venues across diverse domains published after 2025. Our experiments reveal a clear trade-off: while both ClaudeCode and Codex improve with model advances, ClaudeCode achieves higher presentation quality at the cost of more than 10 hallucinations per paper on average, whereas Codex produces fewer hallucinations but lower presentation quality. This work takes a first step toward establishing evaluation frameworks for AI-driven paper writing and improving the understanding of its risks within the research community.

Haruka Kumagai, Leslie Wöhler, Satoshi Ikehata et al.

In response to rising societal awareness of privacy concerns, face anonymization techniques have advanced, including the emergence of face-swapping methods that replace one identity with another. Achieving a balance between anonymity and naturalness in face swapping requires careful selection of identities: overly similar faces compromise anonymity, while dissimilar ones reduce naturalness. Existing models, however, focus on binary identity classification "the same person or not", making it difficult to measure nuanced similarities such as "completely different" versus "highly similar but different." This paper proposes a human-perception-based face similarity metric, creating a dataset of 6,400 triplet annotations and metric learning to predict the similarity. Experimental results demonstrate significant improvements in both face similarity prediction and attribute-based face classification tasks over existing methods.

Jeonghun Baek, Akiko Aizawa, Kiyoharu Aizawa

Large Multimodal Models (LMMs) have demonstrated strong performance in English, but their effectiveness in Japanese remains limited due to the lack of high-quality training data. Current Japanese LMMs often rely on translated English datasets, restricting their ability to capture Japan-specific cultural knowledge. To address this, we explore the potential of Japanese PDF data as a training resource, an area that remains largely underutilized. We introduce a fully automated pipeline that leverages pretrained models to extract image-text pairs from PDFs through layout analysis, OCR, and vision-language pairing, removing the need for manual annotation. Additionally, we construct instruction data from extracted image-text pairs to enrich the training data. To evaluate the effectiveness of PDF-derived data, we train Japanese LMMs and assess their performance on the Japanese LMM Benchmark. Our results demonstrate substantial improvements, with performance gains ranging from 2.1% to 13.8% on Heron-Bench. Further analysis highlights the impact of PDF-derived data on various factors, such as model size and language models, reinforcing its value as a multimodal resource for Japanese LMMs.

Jiafeng Mao, Xueting Wang, Kiyoharu Aizawa

Diffusion models have the ability to generate high quality images by denoising pure Gaussian noise images. While previous research has primarily focused on improving the control of image generation through adjusting the denoising process, we propose a novel direction of manipulating the initial noise to control the generated image. Through experiments on stable diffusion, we show that blocks of pixels in the initial latent images have a preference for generating specific content, and that modifying these blocks can significantly influence the generated image. In particular, we show that modifying a part of the initial image affects the corresponding region of the generated image while leaving other regions unaffected, which is useful for repainting tasks. Furthermore, we find that the generation preferences of pixel blocks are primarily determined by their values, rather than their position. By moving pixel blocks with a tendency to generate user-desired content to user-specified regions, our approach achieves state-of-the-art performance in layout-to-image generation. Our results highlight the flexibility and power of initial image manipulation in controlling the generated image. Project Page: https://ut-mao.github.io/swap.github.io/

Miao Cao, Satoshi Ikehata, Kiyoharu Aizawa

360° cameras have gained popularity over the last few years. In this paper, we propose two fundamental techniques -- Field-of-View IoU (FoV-IoU) and 360Augmentation for object detection in 360° images. Although most object detection neural networks designed for the perspective images are applicable to 360° images in equirectangular projection (ERP) format, their performance deteriorates owing to the distortion in ERP images. Our method can be readily integrated with existing perspective object detectors and significantly improves the performance. The FoV-IoU computes the intersection-over-union of two Field-of-View bounding boxes in a spherical image which could be used for training, inference, and evaluation while 360Augmentation is a data augmentation technique specific to 360° object detection task which randomly rotates a spherical image and solves the bias due to the sphere-to-plane projection. We conduct extensive experiments on the 360indoor dataset with different types of perspective object detectors and show the consistent effectiveness of our method.

Jiafeng Mao, Qing Yu, Yoko Yamakata et al.

Supervised training of object detectors requires well-annotated large-scale datasets, whose production is costly. Therefore, some efforts have been made to obtain annotations in economical ways, such as cloud sourcing. However, datasets obtained by these methods tend to contain noisy annotations such as inaccurate bounding boxes and incorrect class labels. In this study, we propose a new problem setting of training object detectors on datasets with entangled noises of annotations of class labels and bounding boxes. Our proposed method efficiently decouples the entangled noises, corrects the noisy annotations, and subsequently trains the detector using the corrected annotations. We verified the effectiveness of our proposed method and compared it with the baseline on noisy datasets with different noise levels. The experimental results show that our proposed method significantly outperforms the baseline.

Jinjin Gu, Haoming Cai, Chao Dong et al.

This paper reports on the NTIRE 2021 challenge on perceptual image quality assessment (IQA), held in conjunction with the New Trends in Image Restoration and Enhancement workshop (NTIRE) workshop at CVPR 2021. As a new type of image processing technology, perceptual image processing algorithms based on Generative Adversarial Networks (GAN) have produced images with more realistic textures. These output images have completely different characteristics from traditional distortions, thus pose a new challenge for IQA methods to evaluate their visual quality. In comparison with previous IQA challenges, the training and testing datasets in this challenge include the outputs of perceptual image processing algorithms and the corresponding subjective scores. Thus they can be used to develop and evaluate IQA methods on GAN-based distortions. The challenge has 270 registered participants in total. In the final testing stage, 13 participating teams submitted their models and fact sheets. Almost all of them have achieved much better results than existing IQA methods, while the winning method can demonstrate state-of-the-art performance.

Daiki Tanaka, Daiki Ikami, Kiyoharu Aizawa

Positive-unlabeled learning refers to the process of training a binary classifier using only positive and unlabeled data. Although unlabeled data can contain positive data, all unlabeled data are regarded as negative data in existing positive-unlabeled learning methods, which resulting in diminishing performance. We provide a new perspective on this problem -- considering unlabeled data as noisy-labeled data, and introducing a new formulation of PU learning as a problem of joint optimization of noisy-labeled data. This research presents a methodology that assigns initial pseudo-labels to unlabeled data which is used as noisy-labeled data, and trains a deep neural network using the noisy-labeled data. Experimental results demonstrate that the proposed method significantly outperforms the state-of-the-art methods on several benchmark datasets.

Naoki Sugimoto, Yoshihito Ebine, Kiyoharu Aizawa

We propose a new Movie Map system, with an interface for exploring cities. The system consists of four stages; acquisition, analysis, management, and interaction. In the acquisition stage, omnidirectional videos are taken along streets in target areas. Frames of the video are localized on the map, intersections are detected, and videos are segmented. Turning views at intersections are subsequently generated. By connecting the video segments following the specified movement in an area, we can view the streets better. The interface allows for easy exploration of a target area, and it can show virtual billboards of stores in the view. We conducted user studies to compare our system to the GSV in a scenario where users could freely move and explore to find a landmark. The experiment showed that our system had a better user experience than GSV.

Haruka Aoki, Koki Tsubota, Hikaru Ikuta et al.

Designing fonts for languages with a large number of characters, such as Japanese and Chinese, is an extremely labor-intensive and time-consuming task. In this study, we addressed the problem of automatically generating Japanese typographic fonts from only a few font samples, where the synthesized glyphs are expected to have coherent characteristics, such as skeletons, contours, and serifs. Existing methods often fail to generate fine glyph images when the number of style reference glyphs is extremely limited. Herein, we proposed a simple but powerful framework for extracting better style features. This framework introduces deep metric learning to style encoders. We performed experiments using black-and-white and shape-distinctive font datasets and demonstrated the effectiveness of the proposed framework.

Takumi Kawashima, Qing Yu, Akari Asai et al.

We propose a new optimization framework for aleatoric uncertainty estimation in regression problems. Existing methods can quantify the error in the target estimation, but they tend to underestimate it. To obtain the predictive uncertainty inherent in an observation, we propose a new separable formulation for the estimation of a signal and of its uncertainty, avoiding the effect of overfitting. By decoupling target estimation and uncertainty estimation, we also control the balance between signal estimation and uncertainty estimation. We conduct three types of experiments: regression with simulation data, age estimation, and depth estimation. We demonstrate that the proposed method outperforms a state-of-the-art technique for signal and uncertainty estimation.

Daichi Horita, Kiyoharu Aizawa

There are five features to consider when using generative adversarial networks to apply makeup to photos of the human face. These features include (1) facial components, (2) interactive color adjustments, (3) makeup variations, (4) robustness to poses and expressions, and the (5) use of multiple reference images. Several related works have been proposed, mainly using generative adversarial networks (GAN). Unfortunately, none of them have addressed all five features simultaneously. This paper closes the gap with an innovative style- and latent-guided GAN (SLGAN). We provide a novel, perceptual makeup loss and a style-invariant decoder that can transfer makeup styles based on histogram matching to avoid the identity-shift problem. In our experiments, we show that our SLGAN is better than or comparable to state-of-the-art methods. Furthermore, we show that our proposal can interpolate facial makeup images to determine the unique features, compare existing methods, and help users find desirable makeup configurations.