When a model is trying to gather information in an interactive setting, it benefits from asking informative questions. However, in the case of a grounded multi-turn image identification task, previous studies have been constrained to polar yes/no questions, limiting how much information the model can gain in a single turn. We present an approach that formulates more informative, open-ended questions. In doing so, we discover that off-the-shelf visual question answering (VQA) models often make presupposition errors, which standard information gain question selection methods fail to account for. To address this issue, we propose a method that can incorporate presupposition handling into both question selection and belief updates. Specifically, we use a two-stage process, where the model first filters out images which are irrelevant to a given question, then updates its beliefs about which image the user intends. Through self-play and human evaluations, we show that our method is successful in asking informative open-ended questions, increasing accuracy over the past state-of-the-art by 14%, while resulting in 48% more efficient games in human evaluations.
Multimodal Large Language Models (MLLMs) have achieved remarkable success in vision-language tasks but their remote sensing (RS) counterpart are relatively under explored. Unlike natural images, RS imagery presents unique challenges that current MLLMs struggle to handle, particularly in visual grounding and spatial reasoning. This study investigates the limitations of CLIP-based MLLMs in RS, highlighting their failure to differentiate visually distinct yet semantically similar RS images. To address this, we introduce a remote sensing multimodal visual patterns (RSMMVP) benchmark. It is designed to evaluate MLLMs in RS tasks by identifying the CLIP-blind pairs, where CLIP-based models incorrectly assign high similarity scores to visually distinct RS images. Through a visual question answering (VQA) evaluation, we analyze the performance of state-of-the-art MLLMs, revealing significant limitations in RS specific representation learning. The results provide valuable insights into the weaknesses of CLIP-based visual encoding and offer a foundation for future research to develop more effective MLLMs tailored for remote sensing applications.
Multimodal Large Language Models (MLLMs) have recently received substantial interest, which shows their emerging potential as general-purpose models for various vision-language tasks. MLLMs involve significant external knowledge within their parameters; however, it is challenging to continually update these models with the latest knowledge, which involves huge computational costs and poor interpretability. Retrieval augmentation techniques have proven to be effective plugins for both LLMs and MLLMs. In this study, we propose multimodal adaptive Retrieval-Augmented Bootstrapping Language-Image Pre-training (RA-BLIP), a novel retrieval-augmented framework for various MLLMs. Considering the redundant information within vision modality, we first leverage the question to instruct the extraction of visual information through interactions with one set of learnable queries, minimizing irrelevant interference during retrieval and generation. Besides, we introduce a pre-trained multimodal adaptive fusion module to achieve question text-to-multimodal retrieval and integration of multimodal knowledge by projecting visual and language modalities into a unified semantic space. Furthermore, we present an Adaptive Selection Knowledge Generation (ASKG) strategy to train the generator to autonomously discern the relevance of retrieved knowledge, which realizes excellent denoising performance. Extensive experiments on open multimodal question-answering datasets demonstrate that RA-BLIP achieves significant performance and surpasses the state-of-the-art retrieval-augmented models.
Recent progress in VLMs has demonstrated impressive capabilities across a variety of tasks in the natural image domain. Motivated by these advancements, the remote sensing community has begun to adopt VLMs for remote sensing vision-language tasks, including scene understanding, image captioning, and visual question answering. However, existing remote sensing VLMs typically rely on closed-set scene understanding and focus on generic scene descriptions, yet lack the ability to incorporate external knowledge. This limitation hinders their capacity for semantic reasoning over complex or context-dependent queries that involve domain-specific or world knowledge. To address these challenges, we first introduced a multimodal Remote Sensing World Knowledge (RSWK) dataset, which comprises high-resolution satellite imagery and detailed textual descriptions for 14,141 well-known landmarks from 175 countries, integrating both remote sensing domain knowledge and broader world knowledge. Building upon this dataset, we proposed a novel Remote Sensing Retrieval-Augmented Generation (RS-RAG) framework, which consists of two key components. The Multi-Modal Knowledge Vector Database Construction module encodes remote sensing imagery and associated textual knowledge into a unified vector space. The Knowledge Retrieval and Response Generation module retrieves and re-ranks relevant knowledge based on image and/or text queries, and incorporates the retrieved content into a knowledge-augmented prompt to guide the VLM in producing contextually grounded responses. We validated the effectiveness of our approach on three representative vision-language tasks, including image captioning, image classification, and visual question answering, where RS-RAG significantly outperformed state-of-the-art baselines.
The rapid advancement of Large Vision-Language models (LVLMs) has demonstrated a spectrum of emergent capabilities. Nevertheless, current models only focus on the visual content of a single scenario, while their ability to associate instances across different scenes has not yet been explored, which is essential for understanding complex visual content, such as movies with multiple characters and intricate plots. Towards movie understanding, a critical initial step for LVLMs is to unleash the potential of character identities memory and recognition across multiple visual scenarios. To achieve the goal, we propose visual instruction tuning with ID reference and develop an ID-Aware Large Vision-Language Model, IDA-VLM. Furthermore, our research introduces a novel benchmark MM-ID, to examine LVLMs on instance IDs memory and recognition across four dimensions: matching, location, question-answering, and captioning. Our findings highlight the limitations of existing LVLMs in recognizing and associating instance identities with ID reference. This paper paves the way for future artificial intelligence systems to possess multi-identity visual inputs, thereby facilitating the comprehension of complex visual narratives like movies.
Modern deep learning models have demonstrated outstanding performance on discovering the underlying mechanisms when both visual appearance and intrinsic relations (e.g., causal structure) data are sufficient, such as Disentangled Representation Learning (DRL), Causal Representation Learning (CRL) and Visual Question Answering (VQA) methods. However, generalization ability of these models is challenged when the visual domain shifts and the relations data is absent during finetuning. To address this challenge, we propose a novel learning framework, Look, Learn and Leverage (L$^3$), which decomposes the learning process into three distinct phases and systematically utilize the class-agnostic segmentation masks as the common symbolic space to align visual domains. Thus, a relations discovery model can be trained on the source domain, and when the visual domain shifts and the intrinsic relations are absent, the pretrained relations discovery model can be directly reused and maintain a satisfactory performance. Extensive performance evaluations are conducted on three different tasks: DRL, CRL and VQA, and show outstanding results on all three tasks, which reveals the advantages of L$^3$.
An effective method for combining frozen large language models (LLM) and visual encoders involves a resampler module that creates a `visual prompt' which is provided to the LLM, along with the textual prompt. While this approach has enabled impressive performance across many coarse-grained tasks like image captioning and visual question answering, more fine-grained tasks that require spatial understanding have not been thoroughly examined. In this paper, we use \textit{diagnostic classifiers} to measure the extent to which the visual prompt produced by the resampler encodes spatial information. Our results show that this information is largely absent from the resampler output when kept frozen during training of the classifiers. However, when the resampler and classifier are trained jointly, we observe a significant performance boost. This shows that the compression achieved by the resamplers can in principle encode the requisite spatial information, but that more object-aware objectives are needed at the pretraining stage to facilitate this capability
건설 모니터링에 사용되는 컴퓨터 비전(CV) 시스템은 불리한 환경 및 시각적 조건에서 성능이 저하될 수 있으나, 기존 건설 이미지 데이터셋에서는 이러한 조건이 여전히 과소 대표되고 있습니다. 본 논문에서는 3,109개의 실제 소스 장면에서 파생된 34,199개의 이미지를 포함하는 쌍(paired) 형태의 합성 건설 현장 이미지 데이터셋인 ConSynth-X를 제시합니다. 이 데이터셋은 강수, 안개, 야간 조명, 야간 악천후, 소형 객체 또는 원거리 뷰를 아우르는 11개의 조건별 하위 집합으로 구성됩니다. 각 합성 이미지는 해당하는 소스 장면에 연결되어 있어 환경 및 시각적 조건 전반에 걸친 통제된 비교를 가능하게 합니다. ConSynth-X는 소스 기반 주석, 생성 메타데이터, 출처(provenance) 정보 및 이미지 품질 지표를 포함하여 객체 검출(object detection), 이미지 캡셔닝, 시각적 그라운딩(visual grounding), 시각적 질의응답(VQA)을 지원합니다. 기술적 검증(validation)에서는 임베딩(embedding) 기반 유사도 및 분포 분석을 사용하여 소스-합성 충실도와 실제 악조건 이미지와의 정렬(alignment)을 평가합니다. 이 데이터셋은 까다로운 현장 조건에서 건설 비전 및 비전-언어 모델의 강건성을 평가하고 개선하기 위한 체계적인 자원을 제공합니다.
We present a simple dynamic batching approach applicable to a large class of dynamic architectures that consistently yields speedups of over 10x. We provide performance bounds when the architecture is not known a priori and a stronger bound in the special case where the architecture is a predetermined balanced tree. We evaluate our approach on Johnson et al.'s recent visual question answering (VQA) result of his CLEVR dataset by Inferring and Executing Programs (IEP). We also evaluate on sparsely gated mixture of experts layers and achieve speedups of up to 1000x over the naive implementation.
Representations of sets are challenging to learn because operations on sets should be permutation-invariant. To this end, we propose a Permutation-Optimisation module that learns how to permute a set end-to-end. The permuted set can be further processed to learn a permutation-invariant representation of that set, avoiding a bottleneck in traditional set models. We demonstrate our model's ability to learn permutations and set representations with either explicit or implicit supervision on four datasets, on which we achieve state-of-the-art results: number sorting, image mosaics, classification from image mosaics, and visual question answering.
To combat COVID-19, both clinicians and scientists need to digest vast amounts of relevant biomedical knowledge in scientific literature to understand the disease mechanism and related biological functions. We have developed a novel and comprehensive knowledge discovery framework, COVID-KG to extract fine-grained multimedia knowledge elements (entities and their visual chemical structures, relations, and events) from scientific literature. We then exploit the constructed multimedia knowledge graphs (KGs) for question answering and report generation, using drug repurposing as a case study. Our framework also provides detailed contextual sentences, subfigures, and knowledge subgraphs as evidence.
Vision language models (VLMs) demonstrate impressive capabilities in visual question answering and image captioning, acting as a crucial link between visual and language models. However, existing open-source VLMs heavily rely on pretrained and frozen vision encoders (such as CLIP). Despite CLIP's robustness across diverse domains, it still exhibits non-negligible image understanding errors. These errors propagate to the VLM responses, resulting in sub-optimal performance. In our work, we propose an efficient and robust method for updating vision encoders within VLMs. Our approach selectively and locally updates encoders, leading to substantial performance improvements on data where previous mistakes occurred, while maintaining overall robustness. Furthermore, we demonstrate the effectiveness of our method during continual few-shot updates. Theoretical grounding, generality, and computational efficiency characterize our approach.
Charts are very popular for analyzing data, visualizing key insights and answering complex reasoning questions about data. To facilitate chart-based data analysis using natural language, several downstream tasks have been introduced recently such as chart question answering and chart summarization. However, most of the methods that solve these tasks use pretraining on language or vision-language tasks that do not attempt to explicitly model the structure of the charts (e.g., how data is visually encoded and how chart elements are related to each other). To address this, we first build a large corpus of charts covering a wide variety of topics and visual styles. We then present UniChart, a pretrained model for chart comprehension and reasoning. UniChart encodes the relevant text, data, and visual elements of charts and then uses a chart-grounded text decoder to generate the expected output in natural language. We propose several chart-specific pretraining tasks that include: (i) low-level tasks to extract the visual elements (e.g., bars, lines) and data from charts, and (ii) high-level tasks to acquire chart understanding and reasoning skills. We find that pretraining the model on a large corpus with chart-specific low- and high-level tasks followed by finetuning on three down-streaming tasks results in state-of-the-art performance on three downstream tasks.
Multimodal foundation models (MFMs) have demonstrated significant success in tasks such as visual captioning, question answering, and image-text retrieval. However, these models face inherent limitations due to their finite internal capacity, which restricts their ability to process extended temporal sequences, a crucial requirement for comprehensive video and audio analysis. To overcome these challenges, we introduce a specialized cognitive module, temporal working memory (TWM), which aims to enhance the temporal modeling capabilities of MFMs. It selectively retains task-relevant information across temporal dimensions, ensuring that critical details are preserved throughout the processing of video and audio content. The TWM uses a query-guided attention approach to focus on the most informative multimodal segments within temporal sequences. By retaining only the most relevant content, TWM optimizes the use of the model's limited capacity, enhancing its temporal modeling ability. This plug-and-play module can be easily integrated into existing MFMs. With our TWM, nine state-of-the-art models exhibit significant performance improvements across tasks such as video captioning, question answering, and video-text retrieval. By enhancing temporal modeling, TWM extends the capability of MFMs to handle complex, time-sensitive data effectively. Our code is available at https://github.com/xid32/NAACL_2025_TWM.
Token merging has emerged as an effective strategy to accelerate Vision Transformers (ViT) by reducing computational costs. However, existing methods primarily rely on the visual token's feature similarity for token merging, overlooking the potential of integrating spatial information, which can serve as a reliable criterion for token merging in the early layers of ViT, where the visual tokens only possess weak visual information. In this paper, we propose ToSA, a novel token merging method that combines both semantic and spatial awareness to guide the token merging process. ToSA leverages the depth image as input to generate pseudo spatial tokens, which serve as auxiliary spatial information for the visual token merging process. With the introduced spatial awareness, ToSA achieves a more informed merging strategy that better preserves critical scene structure. Experimental results demonstrate that ToSA outperforms previous token merging methods across multiple benchmarks on visual and embodied question answering while largely reducing the runtime of the ViT, making it an efficient solution for ViT acceleration. The code will be available at: https://github.com/hsiangwei0903/ToSA
Medical visual question answering aims to support clinical decision-making by enabling models to answer natural language questions based on medical images. While recent advances in multi-modal learning have significantly improved performance, current methods still suffer from limited answer reliability and poor interpretability, impairing the ability of clinicians and patients to understand and trust model-generated answers. To address this, this work first proposes a Thinking with Visual Grounding (ThinkVG) dataset wherein the answer generation is decomposed into intermediate reasoning steps that explicitly ground relevant visual regions of the medical image, thereby providing fine-grained explainability. Furthermore, we introduce a novel verifiable reward mechanism for reinforcement learning to guide post-training, improving the alignment between the model's reasoning process and its final answer. Remarkably, our method achieves comparable performance using only one-eighth of the training data, demonstrating the efficiency and effectiveness of the proposal. The dataset is available at https://huggingface.co/datasets/BoKelvin/GEMeX-ThinkVG.
Video Anomaly Detection (VAD) systems can autonomously monitor and identify disturbances, reducing the need for manual labor and associated costs. However, current VAD systems are often limited by their superficial semantic understanding of scenes and minimal user interaction. Additionally, the prevalent data scarcity in existing datasets restricts their applicability in open-world scenarios. In this paper, we introduce Hawk, a novel framework that leverages interactive large Visual Language Models (VLM) to interpret video anomalies precisely. Recognizing the difference in motion information between abnormal and normal videos, Hawk explicitly integrates motion modality to enhance anomaly identification. To reinforce motion attention, we construct an auxiliary consistency loss within the motion and video space, guiding the video branch to focus on the motion modality. Moreover, to improve the interpretation of motion-to-language, we establish a clear supervisory relationship between motion and its linguistic representation. Furthermore, we have annotated over 8,000 anomaly videos with language descriptions, enabling effective training across diverse open-world scenarios, and also created 8,000 question-answering pairs for users' open-world questions. The final results demonstrate that Hawk achieves SOTA performance, surpassing existing baselines in both video description generation and question-answering. Our codes/dataset/demo will be released at https://github.com/jqtangust/hawk.
Medical Visual Question Answering (MedVQA) is a promising field for developing clinical decision support systems, yet progress is often limited by the available datasets, which can lack clinical complexity and visual diversity. To address these gaps, we introduce Kvasir-VQA-x1, a new, large-scale dataset for gastrointestinal (GI) endoscopy. Our work significantly expands upon the original Kvasir-VQA by incorporating 159,549 new question-answer pairs that are designed to test deeper clinical reasoning. We developed a systematic method using large language models to generate these questions, which are stratified by complexity to better assess a model's inference capabilities. To ensure our dataset prepares models for real-world clinical scenarios, we have also introduced a variety of visual augmentations that mimic common imaging artifacts. The dataset is structured to support two main evaluation tracks: one for standard VQA performance and another to test model robustness against these visual perturbations. By providing a more challenging and clinically relevant benchmark, Kvasir-VQA-x1 aims to accelerate the development of more reliable and effective multimodal AI systems for use in clinical settings. The dataset is fully accessible and adheres to FAIR data principles, making it a valuable resource for the wider research community. Code and data: https://github.com/Simula/Kvasir-VQA-x1 and https://huggingface.co/datasets/SimulaMet/Kvasir-VQA-x1
We present ReXVQA, the largest and most comprehensive benchmark for visual question answering (VQA) in chest radiology, comprising approximately 696,000 questions paired with 160,000 chest X-rays studies across training, validation, and test sets. Unlike prior efforts that rely heavily on template based queries, ReXVQA introduces a diverse and clinically authentic task suite reflecting five core radiological reasoning skills: presence assessment, location analysis, negation detection, differential diagnosis, and geometric reasoning. We evaluate eight state-of-the-art multimodal large language models, including MedGemma-4B-it, Qwen2.5-VL, Janus-Pro-7B, and Eagle2-9B. The best-performing model (MedGemma) achieves 83.24% overall accuracy. To bridge the gap between AI performance and clinical expertise, we conducted a comprehensive human reader study involving 3 radiology residents on 200 randomly sampled cases. Our evaluation demonstrates that MedGemma achieved superior performance (83.84% accuracy) compared to human readers (best radiology resident: 77.27%), representing a significant milestone where AI performance exceeds expert human evaluation on chest X-ray interpretation. The reader study reveals distinct performance patterns between AI models and human experts, with strong inter-reader agreement among radiologists while showing more variable agreement patterns between human readers and AI models. ReXVQA establishes a new standard for evaluating generalist radiological AI systems, offering public leaderboards, fine-grained evaluation splits, structured explanations, and category-level breakdowns. This benchmark lays the foundation for next-generation AI systems capable of mimicking expert-level clinical reasoning beyond narrow pathology classification. Our dataset will be open-sourced at https://huggingface.co/datasets/rajpurkarlab/ReXVQA
Neural Module Networks (NMN) are a compelling method for visual question answering, enabling the translation of a question into a program consisting of a series of reasoning sub-tasks that are sequentially executed on the image to produce an answer. NMNs provide enhanced explainability compared to integrated models, allowing for a better understanding of the underlying reasoning process. To improve the effectiveness of NMNs we propose to exploit features obtained by a large-scale cross-modal encoder. Also, the current training approach of NMNs relies on the propagation of module outputs to subsequent modules, leading to the accumulation of prediction errors and the generation of false answers. To mitigate this, we introduce an NMN learning strategy involving scheduled teacher guidance. Initially, the model is fully guided by the ground-truth intermediate outputs, but gradually transitions to an autonomous behavior as training progresses. This reduces error accumulation, thus improving training efficiency and final performance.We demonstrate that by incorporating cross-modal features and employing more effective training techniques for NMN, we achieve a favorable balance between performance and transparency in the reasoning process.