Visual Question Research Articles

Automated construction safety reporting system integrating deep learning-based real-time advanced detection and visual question answering

The construction sector is globally acknowledged as one of the most hazardous industries, owing to the vulnerability of its workers to accidents, injuries, and even loss of life. Effective precautionary measures are necessary and ensuring the use of personal protective equipment (PPE) by workers is crucial for protecting them from accidents. Existing deep learning-based PPE detection systems mainly use simple vision-based target detection methods for tasks such as the identification of helmets or vests, and they tend to be task-specific. However, the identification of specific PPE based on respective job types and maintaining detailed safety records, requires further innovative approaches. In this paper, we propose an innovative intelligent system that not only accurately recognizes specific PPE according to the needs of different work types but also automatically generates safety inspection reports and establishes complete safety records, thus providing critical data to support accident investigations. The proposed system integrates a target detection model, visual question answering model, and text-based analysis of the relevant regulations to realize real-time detection of PPE and automatic generation of safety inspection reports. The experimental results show that the proposed YOLOv8n-DCA network strikes a good balance between performance and computational cost—, with a mAP value of 86%. Compared to the original YOLOv8n network, the mAP value is improved by 5.1%, while the model parameters and size are significantly reduced. Further, the visual question answering model exhibited a precision is 95.9. Finally, the automatic generation of safety inspection reports was successfully realized, verifying the feasibility of the developed system. This innovative system promises a comprehensive and efficient PPE management solution for the construction industry to ensure worker safety and provide strong data support.

OphGLM: An ophthalmology large language-and-vision assistant

Vision computer-aided diagnostic methods have been used in early ophthalmic disease screening and diagnosis. However, the limited output formats of these methods lead to poor human–computer interaction and low clinical applicability value. Thus, ophthalmic visual question answering is worth studying. Unfortunately, no practical solutions exist before Large Language Models(LLMs). In this paper, we investigate the ophthalmic visual diagnostic interaction problem. We construct an ophthalmology large language-and-vision assistant, OphGLM, consisting of an image encoder, a text encoder, a fusion module, and an LLM module. We establish a new Chinese ophthalmic fine-tuning dataset, FundusTuning-CN, including the fundus instruction and conversation sets. Based on FundusTuning-CN, we establish a novel LLM-tuning strategy to introduce visual model understanding and ophthalmic knowledge into LLMs at a low cost and high efficiency. Leveraging the pre-training of the image encoder, OphGLM demonstrates strong visual understanding and surpasses open-source visual language models in common fundus disease classification tasks. The FundusTuning-CN enables OphGLM to surpass open-source medical LLMs in both ophthalmic knowledge and interactive capabilities. Our proposed OphGLM has the potential to revolutionize clinical applications in ophthalmology. The dataset, code, and models will be publicly available at https://github.com/ML-AILab/OphGLM.

Deep Fuzzy Multiteacher Distillation Network for Medical Visual Question Answering

Deep Fuzzy Multiteacher Distillation Network for Medical Visual Question Answering

Histopathology in focus: a review on explainable multi-modal approaches for breast cancer diagnosis.

Precision and timeliness in breast cancer detection are paramount for improving patient outcomes. Traditional diagnostic methods have predominantly relied on unimodal approaches, but recent advancements in medical data analytics have enabled the integration of diverse data sources beyond conventional imaging techniques. This review critically examines the transformative potential of integrating histopathology images with genomic data, clinical records, and patient histories to enhance diagnostic accuracy and comprehensiveness in multi-modal diagnostic techniques. It explores early, intermediate, and late fusion methods, as well as advanced deep multimodal fusion techniques, including encoder-decoder architectures, attention-based mechanisms, and graph neural networks. An overview of recent advancements in multimodal tasks such as Visual Question Answering (VQA), report generation, semantic segmentation, and cross-modal retrieval is provided, highlighting the utilization of generative AI and visual language models. Additionally, the review delves into the role of Explainable Artificial Intelligence (XAI) in elucidating the decision-making processes of sophisticated diagnostic algorithms, emphasizing the critical need for transparency and interpretability. By showcasing the importance of explainability, we demonstrate how XAI methods, including Grad-CAM, SHAP, LIME, trainable attention, and image captioning, enhance diagnostic precision, strengthen clinician confidence, and foster patient engagement. The review also discusses the latest XAI developments, such as X-VARs, LeGrad, LangXAI, LVLM-Interpret, and ex-ILP, to demonstrate their potential utility in multimodal breast cancer detection, while identifying key research gaps and proposing future directions for advancing the field.

Comparative performance of artificial intelligence models in rheumatology board-level questions: evaluating Google Gemini and ChatGPT-4o.

This study evaluates the performance of AI models, ChatGPT-4o and Google Gemini, in answering rheumatology board-level questions, comparing their effectiveness, reliability, and applicability in clinical practice. A cross-sectional study was conducted using 420 rheumatology questions from the BoardVitals question bank, excluding 27 visual data questions. Both artificial intelligence models categorized the questions according to difficulty (easy, medium, hard) and answered them. In addition, the reliability of the answers was assessed by asking the questions a second time. The accuracy, reliability, and difficulty categorization of the AI models' response to the questions were analyzed. ChatGPT-4o answered 86.9% of the questions correctly, significantly outperforming Google Gemini's 60.2% accuracy (p < 0.001). When the questions were asked a second time, the success rate was 86.7% for ChatGPT-4o and 60.5% for Google Gemini. Both models mainly categorized questions as medium difficulty. ChatGPT-4o showed higher accuracy in various rheumatology subfields, notably in Basic and Clinical Science (p = 0.028), Osteoarthritis (p = 0.023), and Rheumatoid Arthritis (p < 0.001). ChatGPT-4o significantly outperformed Google Gemini in rheumatology board-level questions. This demonstrates the success of ChatGPT-4o in situations requiring complex and specialized knowledge related to rheumatological diseases. The performance of both AI models decreased as the question difficulty increased. This study demonstrates the potential of AI in clinical applications and suggests that its use as a tool to assist clinicians may improve healthcare efficiency in the future. Future studies using real clinical scenarios and real board questions are recommended. Key Points •ChatGPT-4o significantly outperformed Google Gemini in answering rheumatology board-level questions, achieving 86.9% accuracy compared to Google Gemini's 60.2%. •For both AI models, the correct answer rate decreased as the question difficulty increased. •The study demonstrates the potential for AI models to be used in clinical practice as a tool to assist clinicians and improve healthcare efficiency.

Integrating IoT and visual question answering in smart cities: Enhancing educational outcomes

Emerging as a paradigmatic shift in urban development, smart cities harness the potential of advanced information and communication technologies to seamlessly integrate urban functions, optimize resource allocation, and improve the effectiveness of city management. Within the domain of smart education, the imperative application of Visual Question Answering (VQA) technology encounters significant limitations at the prevailing stage, particularly the absence of a robust Internet of Things (IoT) framework and the inadequate incorporation of large pre-trained language models (LLMs) within contemporary smart education paradigms, especially in addressing zero-shot VQA scenarios, which pose considerable challenges. In response to these constraints, this paper introduces an IoT-based smart city framework that is designed to refine the functionality and efficacy of educational systems. This framework is delineated into four cardinal layers: the data collection layer, data transmission layer, data management layer, and application layer. Furthermore, we introduce the innovative TeachVQA methodology at the application layer, synergizing VQA technology with extensive pre-trained language models, thereby considerably enhancing the dissemination and assimilation of educational content. Evaluative metrics in the VQAv2 and OKVQA datasets substantiate that the TeachVQA methodology not only outperforms existing VQA approaches, but also underscores its profound potential and practical relevance in the educational sector.

Adversarial Sample Synthesis for Visual Question Answering

Language prior is a major block to improving the generalization of visual question answering (VQA) models. Recent work has revealed that synthesizing extra training samples to balance training sets is a promising way to alleviate language priors. However, most existing methods synthesize extra samples in a manner independent of training processes, which neglect the fact that the language priors memorized by VQA models are changing during training, resulting in insufficient synthesized samples. In this paper, we propose an adversarial sample synthesis method, which synthesizes different adversarial samples by adversarial masking at different training epochs to cope with the changing memorized language priors. The basic idea behind our method is to use adversarial masking to synthesize adversarial samples that will cause the model to make wrong answers. To this end, we design a generative module to carry out adversarial masking by attacking the VQA model, and introduce a bias-oriented objective to supervise the training of the generative module. We couple the sample synthesis with the training process of the VQA model, which ensures that the synthesized samples at different training epochs are beneficial to the VQA model. We incorporated the proposed method into three VQA models including UpDn, LMH and LXMERT, and conducted experiments on three datasets including VQA-CP v1, VQA-CP v2 and VQA v2. Experimental results demonstrate that a large improvement of our method, such as 16.22% gains on LXMERT in the overall accuracy of VQA-CP v2.

Learning to enhance areal video captioning with visual question answering

ABSTRACT The utilization of Unmanned Aerial Vehicles (UAV) in remote sensing (RS) has witnessed a significant surge, offering valuable insights into Earth dynamics and human activities. However, this has led to a substantial increase in the volume of video data, rendering manual screening and analysis impractical. Consequently, there is a pressing need for the development of automated interpretation models for these aerial videos. In this paper, we propose a novel approach that leverages visual dialogue to enhance aerial video captioning. Our model adopts an encoder-decoder architecture, integrating a Visual Question Answering (VQA) task before the captioning task. The VQA task aims to enrich the captioning process by soliciting additional information about the image content. Specifically, our video encoder utilizes ViT-L/16, while the decoder employs Generative Pre-trained Transformer-2 (Distill-GPT-2). To validate our model, we introduce a novel benchmark dataset named CapERA-VQA, comprising videos accompanied by sets of questions, answers, and captions. Through experimental validation, we demonstrate the effectiveness of our proposed approach in enhancing the automated captioning of aerial videos.

Evaluating text and visual diagnostic capabilities of large language models on questions related to the Breast Imaging Reporting and Data System Atlas 5th edition.

This study aimed to evaluate the performance of large language models (LLMs) and multimodal LLMs in interpreting the Breast Imaging Reporting and Data System (BI-RADS) categories and providing clinical management recommendations for breast radiology in text-based and visual questions. This cross-sectional observational study involved two steps. In the first step, we compared ten LLMs (namely ChatGPT 4o, ChatGPT 4, ChatGPT 3.5, Google Gemini 1.5 Pro, Google Gemini 1.0, Microsoft Copilot, Perplexity, Claude 3.5 Sonnet, Claude 3 Opus, and Claude 3 Opus 200K), general radiologists, and a breast radiologist using 100 text-based multiple-choice questions (MCQs) related to the BI-RADS Atlas 5th edition. In the second step, we assessed the performance of five multimodal LLMs (ChatGPT 4o, ChatGPT 4V, Claude 3.5 Sonnet, Claude 3 Opus, and Google Gemini 1.5 Pro) in assigning BI-RADS categories and providing clinical management recommendations on 100 breast ultrasound images. The comparison of correct answers and accuracy by question types was analyzed using McNemar's and chi-squared tests. Management scores were analyzed using the Kruskal- Wallis and Wilcoxon tests. Claude 3.5 Sonnet achieved the highest accuracy in text-based MCQs (90%), followed by ChatGPT 4o (89%), outperforming all other LLMs and general radiologists (78% and 76%) (P < 0.05), except for the Claude 3 Opus models and the breast radiologist (82%) (P > 0.05). Lower-performing LLMs included Google Gemini 1.0 (61%) and ChatGPT 3.5 (60%). Performance across different categories of showed no significant variation among LLMs or radiologists (P > 0.05). For breast ultrasound images, Claude 3.5 Sonnet achieved 59% accuracy, significantly higher than other multimodal LLMs (P < 0.05). Management recommendations were evaluated using a 3-point Likert scale, with Claude 3.5 Sonnet scoring the highest (mean: 2.12 ± 0.97) (P < 0.05). Accuracy varied significantly across BI-RADS categories, except Claude 3 Opus (P < 0.05). Gemini 1.5 Pro failed to answer any BI-RADS 5 questions correctly. Similarly, ChatGPT 4V failed to answer any BI-RADS 1 questions correctly, making them the least accurate in these categories (P < 0.05). Although LLMs such as Claude 3.5 Sonnet and ChatGPT 4o show promise in text-based BI-RADS assessments, their limitations in visual diagnostics suggest they should be used cautiously and under radiologists' supervision to avoid misdiagnoses. This study demonstrates that while LLMs exhibit strong capabilities in text-based BI-RADS assessments, their visual diagnostic abilities are currently limited, necessitating further development and cautious application in clinical practice.

Prompting Large Language Models with Knowledge-Injection for Knowledge-Based Visual Question Answering

Prompting Large Language Models with Knowledge-Injection for Knowledge-Based Visual Question Answering

Fair Attention Network for Robust Visual Question Answering

Fair Attention Network for Robust Visual Question Answering

Robust-EQA: Robust Learning for Embodied Question Answering With Noisy Labels.

Embodied question answering (EQA) is a recently emerged research field in which an agent is asked to answer the user's questions by exploring the environment and collecting visual information. Plenty of researchers turn their attention to the EQA field due to its broad potential application areas, such as in-home robots, self-driven mobile, and personal assistants. High-level visual tasks, such as EQA, are susceptible to noisy inputs, because they have complex reasoning processes. Before the profits of the EQA field can be applied to practical applications, good robustness against label noise needs to be equipped. To tackle this problem, we propose a novel label noise-robust learning algorithm for the EQA task. First, a joint training co-regularization noise-robust learning method is proposed for noisy filtering of the visual question answering (VQA) module, which trains two parallel network branches by one loss function. Then, a two-stage hierarchical robust learning algorithm is proposed to filter out noisy navigation labels in both trajectory level and action level. Finally, by taking purified labels as inputs, a joint robust learning mechanism is given to coordinate the work of the whole EQA system. Empirical results demonstrate that, under extremely noisy environments (45% of noisy labels) and low-level noisy environments (20% of noisy labels), the robustness of deep learning models trained by our algorithm is superior to the existing EQA models in noisy environments.

VADS: Visuo-Adaptive DualStrike attack on visual question answer

Visual Question Answering (VQA) is a fundamental task in computer vision and natural language process fields. The adversarial vulnerability of VQA models is crucial for their reliability in real-world applications. However, current VQA attacks are mainly focused on the white-box and transfer-based settings, which require the attacker to have full or partial prior knowledge of victim VQA models. Besides that, query-based VQA attacks require a massive amount of query times, which the victim model may detect. In this paper, we propose the Visuo-Adaptive DualStrike (VADS) attack, a novel adversarial attack method combining transfer-based and query-based strategies to exploit vulnerabilities in VQA systems. Unlike current VQA attacks focusing on either approach, VADS leverages a momentum-like ensemble method to search potential attack targets and compress the perturbation. After that, our method employs a query-based strategy to dynamically adjust the weight of perturbation per surrogate model. We evaluate the effectiveness of VADS across 8 VQA models and two datasets. The results demonstrate that VADS outperforms existing adversarial techniques in both efficiency and success rate. Our code is available at: https://github.com/stevenzhang9577/VADS.

A lightweight Transformer-based visual question answering network with Weight-Sharing Hybrid Attention

Recent advances show that Transformer-based models and object detection-based models play an indispensable role in VQA. However, object detection-based models have significant limitations due to their redundant and complex detection box generation process. In contrast, Visual and Language Pre-training (VLP) models can achieve better performance, but require high computing power. To this end, we present Weight-Sharing Hybrid Attention Network (WHAN), a lightweight Transformer-based VQA model. In WHAN, we replace the object detection network with Transformer encoder and use LoRA to solve the problem that the language model cannot adapt to interrogative sentences. We propose Weight-Sharing Hybrid Attention (WHA) module with parallel residual adapters, which can significantly reduce the trainable parameters of the model and we design DWA and BVA modules that can allow the model to perform attention operations from different scales. Experiments on VQA-v2, COCO-QA, GQA, and CLEVR datasets show that WHAN achieves competitive performance with far fewer trainable parameters.

Overcoming language priors in visual question answering with cumulative learning strategy

The performance of visual question answering(VQA) has witnessed great progress over the last few years. However, many current VQA models tend to rely on superficial linguistic correlations between questions and answers, often failing to sufficiently learn multi-modal knowledge from both vision and language, and thus suffering significant performance drops. To address this issue, the VQA-CP v2.0 dataset was developed to reduce language biases by greedily re-partitioning the distribution of VQA v2.0’s training and test sets. According to the fact that achieving high performance on real-world datasets requires effective learning from minor classes, in this paper we analyze the presence of skewed long-tail distributions in the VQA-CP v2.0 dataset and propose a new ensemble-based parameter-insensitive framework. This framework is built on two representation learning branches and a joint learning block, which are designed to reduce language biases in VQA tasks. Specifically, the representation learning branches can ensure the superior representative ability learned from the major and minor classes. The joint learning block forces the model to initially concentrate on major classes for robust representation and then gradually shifts its focus towards minor classes for classification during the training progress. Experimental results demonstrate that our approach outperforms the state-of-the-art works on the VQA-CP v2.0 dataset without requiring additional annotations. Notably, on the “num” type, our framework exceeds the second-best method (without extra annotations) by 8.64%. Meanwhile, our approach does not sacrifice accuracy performance on the VQA v2.0 dataset compared with the baseline model.

Performance of ChatGPT 4.0 on Japan's National Physical Therapist Examination: A Comprehensive Analysis of Text and Visual Question Handling.

ChatGPT 4.0, a large-scale language model (LLM) developed by OpenAI, has demonstrated the capability to pass Japan's national medical examination and other medical assessments. However, the impact of imaging-based questions and different question types on its performance has not been thoroughly examined. This study evaluated ChatGPT 4.0's performance on Japan's national examination for physical therapists, particularly its ability to handle complex questions involving images and tables. The study also assessed the model's potential in the field of rehabilitation and its performance with Japanese language inputs. The evaluation utilized 1,000 questions from the 54th to 58th national exams for physical therapists in Japan, comprising 160 general questions and 40 practical questions per exam. All questions were input in Japanese and included additional information such as images or tables. The answers generated by ChatGPT were then compared with the official correct answers. ChatGPT's performance was evaluated based on accuracy rates using various criteria: general and practical questions were analyzed with Fisher's exact test, A-type (single correct answer) and X2-type (two correct answers) questions, text-only questions versus questions with images and tables, and different question lengths using Student's t-test. ChatGPT 4.0 met the passing criteria with an overall accuracy of 73.4%. The accuracy rates for general and practical questions were 80.1% and 46.6%, respectively. No significant difference was found between the accuracy rates for A-type (74.3%) and X2-type (67.4%) questions. However, a significant difference was observed between the accuracy rates for text-only questions (80.5%) and questions with images and tables (35.4%). The results indicate that ChatGPT 4.0 satisfies the passing criteria for the national exam and demonstrates adequate knowledge and application skills. However, its performance on practical questions and those with images and tables is lower, indicating areas for improvement. The effective handling of Japanese inputs suggests its potential use in non-English-speaking regions. ChatGPT 4.0 can pass the national examination for physical therapists, particularly with text-based questions. However, improvements are needed for specialized practical questions and those involving images and tables. The model shows promise for supporting clinical rehabilitation and medical education in Japanese-speaking contexts, though further enhancements are required for a comprehensive application.

Overcoming the Limitations of Learning-Based VQA for Counting Questions with Zero-Shot Learning

Visual question answering (VQA) research has garnered increasing attention in recent years. It is considered a visual Turing test because it requires a computer to respond to textual questions based on an image. Expertise in computer vision, natural language processing, knowledge understanding, and reasoning is required to solve the problem of VQA. Most techniques employed for VQA consist of models that are developed to learn the combination of image and question features along with the expected answer. The techniques chosen for image and question feature extraction and combining the features change with each model. This method of teaching a model of the question–answer pattern is ineffective for queries that involve counting and reasoning. This approach also requires considerable resources and large datasets for the training. The general VQA datasets feature a restricted number of items as responses to counting questions ([Formula: see text]), and the distribution of the answers is not uniform. To investigate these issues in VQA, we created synthetic datasets that could be modified to adjust the number of objects in the image and the amount of occlusion. Specifically, a zero-shot learning VQA system was devised for counting-related questions that provide answers by analyzing the output of an object detector and the query keywords. Using synthetic datasets, our model generated 100% correct results. Testing on the benchmark datasets task directed image understanding challenge (TDIUC) and TallyQA-simple indicated that the proposed model matched the performance of the learning-based baseline models. This methodology can be used efficiently for counting VQA questions confined to certain domains when the number of items to be counted is significant.

A generalist vision-language foundation model for diverse biomedical tasks.

Traditional biomedical artificial intelligence (AI) models, designed for specific tasks or modalities, often exhibit limited flexibility in real-world deployment and struggle to utilize holistic information. Generalist AI holds the potential to address these limitations due to its versatility in interpreting different data types and generating tailored outputs for diverse needs. However, existing biomedical generalist AI solutions are typically heavyweight and closed source to researchers, practitioners and patients. Here, we describe BiomedGPT, the first open-source and lightweight vision-language foundation model, designed as a generalist capable of performing various biomedical tasks. BiomedGPT achieved state-of-the-art results in 16 out of 25 experiments while maintaining a computing-friendly model scale. We also conducted human evaluations to assess the capabilities of BiomedGPT in radiology visual question answering, report generation and summarization. BiomedGPT exhibits robust prediction ability with a low error rate of 3.8% in question answering, satisfactory performance with an error rate of 8.3% in writing complex radiology reports, and competitive summarization ability with a nearly equivalent preference score to human experts. Our method demonstrates that effective training with diverse data can lead to more practical biomedical AI for improving diagnosis and workflow efficiency.

Advancing Accuracy in Multimodal Medical Tasks Through Bootstrapped Language-Image Pretraining (BioMedBLIP): Performance Evaluation Study.

Medical image analysis, particularly in the context of visual question answering (VQA) and image captioning, is crucial for accurate diagnosis and educational purposes. Our study aims to introduce BioMedBLIP models, fine-tuned for VQA tasks using specialized medical data sets such as Radiology Objects in Context and Medical Information Mart for Intensive Care-Chest X-ray, and evaluate their performance in comparison to the state of the art (SOTA) original Bootstrapping Language-Image Pretraining (BLIP) model. We present 9 versions of BioMedBLIP across 3 downstream tasks in various data sets. The models are trained on a varying number of epochs. The findings indicate the strong overall performance of our models. We proposed BioMedBLIP for the VQA generation model, VQA classification model, and BioMedBLIP image caption model. We conducted pretraining in BLIP using medical data sets, producing an adapted BLIP model tailored for medical applications. In VQA generation tasks, BioMedBLIP models outperformed the SOTA on the Semantically-Labeled Knowledge-Enhanced (SLAKE) data set, VQA in Radiology (VQA-RAD), and Image Cross-Language Evaluation Forum data sets. In VQA classification, our models consistently surpassed the SOTA on the SLAKE data set. Our models also showed competitive performance on the VQA-RAD and PathVQA data sets. Similarly, in image captioning tasks, our model beat the SOTA, suggesting the importance of pretraining with medical data sets. Overall, in 20 different data sets and task combinations, our BioMedBLIP excelled in 15 (75%) out of 20 tasks. BioMedBLIP represents a new SOTA in 15 (75%) out of 20 tasks, and our responses were rated higher in all 20 tasks (P<.005) in comparison to SOTA models. Our BioMedBLIP models show promising performance and suggest that incorporating medical knowledge through pretraining with domain-specific medical data sets helps models achieve higher performance. Our models thus demonstrate their potential to advance medical image analysis, impacting diagnosis, medical education, and research. However, data quality, task-specific variability, computational resources, and ethical considerations should be carefully addressed. In conclusion, our models represent a contribution toward the synergy of artificial intelligence and medicine. We have made BioMedBLIP freely available, which will help in further advancing research in multimodal medical tasks.

Robust Visual Question Answering: Datasets, Methods, and Future Challenges.

Visual question answering requires a system to provide an accurate natural language answer given an image and a natural language question. However, it is widely recognized that previous generic VQA methods often tend to memorize biases present in the training data rather than learning proper behaviors, such as grounding images before predicting answers. Therefore, these methods usually achieve high in-distribution but poor out-of-distribution performance. In recent years, various datasets and debiasing methods have been proposed to evaluate and enhance the VQA robustness, respectively. This paper provides the first comprehensive survey focused on this emerging fashion. Specifically, we first provide an overview of the development process of datasets from in-distribution and out-of-distribution perspectives. Then, we examine the evaluation metrics employed by these datasets. Third, we propose a typology that presents the development process, similarities and differences, robustness comparison, and technical features of existing debiasing methods. Furthermore, we analyze and discuss the robustness of representative vision-and-language pre-training models on VQA. Finally, through a thorough review of the available literature and experimental analysis, we discuss the key areas for future research from various viewpoints.