Visual Question Answering Models Research Articles

LRCN: Layer-residual Co-Attention Networks for Visual Question Answering

Visual Question Answering (VQA) is a multimodal task requiring a collaborative understanding of fine-grained visual concepts and language semantics. The key to VQA research is constructing a framework for modeling inter and intramodal information interactions between intricate modalities. Due to the superior global view and the ability to capture the relationships within multimodal information, the Transformer and its variants have become the prime choice for VQA tasks. Despite this layer-by-layer architecture enabling answer reasoning by optimizing multimodal feature information, the information may still be lost when transferred from lower to higher layers. To solve such an issue, we propose a Layer-Residual Mechanism (LRM), a plug-and-play generic approach that can reduce the computation and memory overhead to almost negligible. By adding a residual straight-through line between adjacent layers that cascades the attention block in-depth, LRM mitigates the vanishing during information transfer, stabilizes training, and endeavors to overcome performance decline in VQA models at deeper layers. To verify the effectiveness and generality of the proposed LRM, we apply it to the Encoder-Decoder structure, the Pure-Stacking structure, and a specifically designed Co-Stacking structure that can simultaneously understand textual features and visual features called Layer-residual Co-Attention Networks (LRCN). Extensive ablation studies and comparative experiments on the benchmark VQA v2 and CLEVR datasets show that LRCN significantly outperforms the original architectures, demonstrating the superior effectiveness and compatibility of LRCN.

Cross-Modal self-supervised vision language pre-training with multiple objectives for medical visual question answering

Medical Visual Question Answering (VQA) is a task that aims to provide answers to questions about medical images, which utilizes both visual and textual information in the reasoning process. The absence of large-scale annotated medical VQA datasets presents a formidable obstacle to training a medical VQA model from scratch in an end-to-end manner. Existing works have been using image captioning dataset in the pre-training stage and fine-tuning to downstream VQA tasks. Following the same paradigm, we use a collection of public medical image captioning datasets to pre-train multimodality models in a self-supervised setup, and fine-tune to downstream medical VQA tasks. In the work, we propose a method that featured with Cross-Modal pre-training with Multiple Objectives (CMMO), which includes masked image modelling, masked language modelling, image-text matching, and image-text contrastive learning. The proposed method is designed to associate the visual features of medical images with corresponding medical concepts in captions, for learning aligned vision and language feature representations, and multi-modal interactions. The experimental results reveal that our proposed CMMO method outperforms state-of-the-art methods on three public medical VQA datasets, showing absolute improvements of 2.6%, 0.9%, and 4.0% on the VQA-RAD, PathVQA, and SLAKE dataset, respectively. We also conduct comprehensive ablation studies to validate our method, and visualize the attention maps which show a strong interpretability. The code and pre-trained weights will be released at https://github.com/pengfeiliHEU/CMMO.

Graph-enhanced visual representations and question-guided dual attention for visual question answering

Visual Question Answering (VQA) has witnessed significant advancements recently, due to the application of deep learning in the field of vision-language research. Most current VQA models focus on merging visual and text features, but it is essential for these models to also consider the relationships between different parts of an image and use question information to highlight important features. This study proposes a method to enhance neighboring image region features and learn question-aware visual representations. First, we construct a region graph to represent spatial relationships between objects in the image. Then, graph convolutional networks (GCNs) are used to propagate information across neighboring regions, enriching each region’s feature representation by integrating contextual information. To capture long-range dependencies, the graph is enhanced with random walk with restart (RWR), enabling multi-hop reasoning across distant regions. Furthermore, a question-aware dual attention mechanism is introduced to further refine region features at both region and feature levels, ensuring that the model emphasizes key regions that are critical for answering the question. The enhanced region representations are then combined with the encoded question to predict an answer. Through extensive experiments on VQA benchmarks, the study demonstrates state-of-the-art performance by leveraging regional dependencies and question guidance. The integration of GCNs and random walks in the graph helps capture contextual information to focus visual attention selectively, resulting in significant improvements over existing methods on VQA 1.0 and VQA 2.0 benchmark datasets.

Robust visual question answering via polarity enhancement and contrast

The Visual Question Answering (VQA) task is an important research direction in the field of artificial intelligence, which requires a model that can simultaneously understand visual images and natural language questions, and answer questions related to images. Recent studies have shown that many Visual Question Answering models rely on statistically regular correlations between questions and answers, which in turn weakens the correlation between visual content and textual information. In this work, we propose an unbiased Visual Question Answering method to solve language priors from the perspective of strengthening the contrast between the correct answer and the positive and negative predictions. We design a new model consisting of two modules with different roles. We input the image and the question corresponding to it into the Answer Visual Attention Modules to generate positive prediction output, and then use a Dual Channels Joint Module to generate negative prediction output with great linguistic prior knowledge. Finally, we input the positive and negative predictions together with the correct answer to our newly designed loss function for training. Our method achieves high performance (61.24%) on the VQA-CP v2 dataset. In addition, most existing debiasing methods improve performance on VQA-CP v2 dataset at the cost of reducing performance on VQA v2 dataset, while our method not only does not reduce the accuracy on VQA v2 dataset. Instead, it improves performance on both datasets mentioned above.

A Picture May Be Worth a Hundred Words for Visual Question Answering

How far can textual representations go in understanding images? In image understanding, effective representations are essential. Deep visual features from object recognition models currently dominate various tasks, especially Visual Question Answering (VQA). However, these conventional features often struggle to capture image details in ways that match human understanding, and their decision processes lack interpretability. Meanwhile, the recent progress in language models suggests that descriptive text could offer a viable alternative. This paper investigated the use of descriptive text as an alternative to deep visual features in VQA. We propose to process description–question pairs rather than visual features, utilizing a language-only Transformer model. We also explored data augmentation strategies to enhance training set diversity and mitigate statistical bias. Extensive evaluation shows that textual representations using approximately a hundred words can effectively compete with deep visual features on both the VQA 2.0 and VQA-CP v2 datasets. Our qualitative experiments further reveal that these textual representations enable clearer investigation of VQA model decision processes, thereby improving interpretability.

HCCL: Hierarchical Counterfactual Contrastive Learning for Robust Visual Question Answering

Despite most state-of-the-art models having achieved amazing performance in Visual Question Answering (VQA) , they usually utilize biases to answer the question. Recently, some studies synthesize counterfactual training samples to help the model to mitigate the biases. However, these synthetic samples need extra annotations and often contain noises. Moreover, these methods simply add synthetic samples to the training data to train the model with the cross-entropy loss, which cannot make the best use of synthetic samples to mitigate the biases. In this article, to mitigate the biases in VQA more effectively, we propose a Hierarchical Counterfactual Contrastive Learning (HCCL) method. Firstly, to avoid introducing noises and extra annotations, our method automatically masks the unimportant features in original pairs to obtain positive samples and create mismatched question-image pairs as negative samples. Then our method uses feature-level and answer-level contrastive learning to make the original sample close to positive samples in the feature space, while away from negative samples in both feature and answer spaces. In this way, the VQA model can learn the robust multimodal features and focus on both visual and language information to produce the answer. Our HCCL method can be adopted in different baselines, and the experimental results on VQA v2, VQA-CP, and GQA-OOD datasets show that our method is effective in mitigating the biases in VQA, which improves the robustness of the VQA model.

A robust visual question answering approach to reduce multimodal bias

Currently, many visual question answering models have bias problems. Specifically, when the question-answer relationship in the training data shows a more obvious mapping relationship, the model shows poor generalization ability. For such biased predictions, existing research work mainly considers language bias, while ignoring the bias information introduced by images. In order to enhance the robustness of visual question answering models, a bias reduction method is proposed, and on this basis, the influence of language and visual information on bias is explored. Furthermore, two bias learning branches are constructed to capture language bias and the bias caused by language and images respectively, and the bias reduction method is used to obtain more robust prediction results. Finally, according to the difference in prediction probability between the standard visual question answering and bias branches, the samples are dynamically weighted, so that the model can dynamically adjust the learning degree for samples with different bias levels. Experiments on datasets such as VQA-CP v2.0 prove the effectiveness of the proposed method and alleviate the influence of bias on the model.

Modular Co-attention Networks in Nepali Visual Question Answering Systems

Visual question answering (VQA) has been regarded as a challenging task requiring a perfect blend of computer vision and natural language processing. As no dataset was available to train such a model for the Nepali language, a new dataset was developed during the research by translating the VQAv2 dataset. Then the dataset consisting of 202,577 images and 886,560 questions was used to train an attention-based VQA model. The dataset consists of yes/no, counting, and other questions with primarily one-word answers. Modular Co-attention Network (MCAN) was applied to the visual features extracted using the Faster RCNN framework and question embeddings extracted using the Nepali GloVe model. After co-attending the visual and language features for a few cascaded MCAN layers, the features are fused to train the whole network. During evaluation, an overall accuracy of 69.87% was obtained with 81.09% accuracy in yes/no type questions. The results surpassed the performance of models developed for Hindi and Bengali languages. Overall, novel research has been done in the Nepali Language VQA domain paving the way for further advancements.

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.

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.

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.

ViCLEVR: a visual reasoning dataset and hybrid multimodal fusion model for visual question answering in Vietnamese

ViCLEVR: a visual reasoning dataset and hybrid multimodal fusion model for visual question answering in Vietnamese

Visual Question Answering Models for Zero-Shot Pedestrian Attribute Recognition: A Comparative Study

Pedestrian Attribute Recognition (PAR) poses a significant challenge in developing automatic systems that enhance visual surveillance and human interaction. In this study, we investigate using Visual Question Answering (VQA) models to address the zero-shot PAR problem. Inspired by the impressive results achieved by a zero-shot VQA strategy during the PAR Contest at the 20th International Conference on Computer Analysis of Images and Patterns in 2023, we conducted a comparative study across three state-of-the-art VQA models, two of them based on BLIP-2 and the third one based on the Plug-and-Play VQA framework. Our analysis focuses on performance, robustness, contextual question handling, processing time, and classification errors. Our findings demonstrate that both BLIP-2-based models are better suited for PAR, with nuances related to the adopted frozen Large Language Model. Specifically, the Open Pre-trained Transformers based model performs well in benchmark color estimation tasks, while FLANT5XL provides better results for the considered binary tasks. In summary, zero-shot PAR based on VQA models offers highly competitive results, with the advantage of avoiding training costs associated with multipurpose classifiers.

Advancing surgical VQA with scene graph knowledge

PurposeThe modern operating room is becoming increasingly complex, requiring innovative intra-operative support systems. While the focus of surgical data science has largely been on video analysis, integrating surgical computer vision with natural language capabilities is emerging as a necessity. Our work aims to advance visual question answering (VQA) in the surgical context with scene graph knowledge, addressing two main challenges in the current surgical VQA systems: removing question–condition bias in the surgical VQA dataset and incorporating scene-aware reasoning in the surgical VQA model design.MethodsFirst, we propose a surgical scene graph-based dataset, SSG-VQA, generated by employing segmentation and detection models on publicly available datasets. We build surgical scene graphs using spatial and action information of instruments and anatomies. These graphs are fed into a question engine, generating diverse QA pairs. We then propose SSG-VQA-Net, a novel surgical VQA model incorporating a lightweight Scene-embedded Interaction Module, which integrates geometric scene knowledge in the VQA model design by employing cross-attention between the textual and the scene features.ResultsOur comprehensive analysis shows that our SSG-VQA dataset provides a more complex, diverse, geometrically grounded, unbiased and surgical action-oriented dataset compared to existing surgical VQA datasets and SSG-VQA-Net outperforms existing methods across different question types and complexities. We highlight that the primary limitation in the current surgical VQA systems is the lack of scene knowledge to answer complex queries.ConclusionWe present a novel surgical VQA dataset and model and show that results can be significantly improved by incorporating geometric scene features in the VQA model design. We point out that the bottleneck of the current surgical visual question–answer model lies in learning the encoded representation rather than decoding the sequence. Our SSG-VQA dataset provides a diagnostic benchmark to test the scene understanding and reasoning capabilities of the model. The source code and the dataset will be made publicly available at: https://github.com/CAMMA-public/SSG-VQA.

Deep learning-based bridge damage cause estimation from multiple images using visual question answering

This paper presents a framework for estimating the cause of damage to bridge members by combining Structure from Motion (SfM) and Visual Question Answering (VQA) techniques. A VQA model was developed that uses bridge images for dataset creation and outputs the damage or member name and its existence based on the images and questions. In the developed model, the correct answer rate for questions requiring the member or damage name were 67.4 and 68.9%, respectively. The correct answer rate for questions requiring a yes/no answer was 99.1%. Based on the developed model, a damage cause estimation method was proposed. In the proposed method, the damage causes are narrowed down by inputting new questions to the VQA model, which are determined based on the surrounding images obtained via SfM and the results of the VQA model. Subsequently, the proposed method was then applied to an actual bridge and shown to be capable of determining damage and estimating its cause. The proposed method could be used to prevent damage causes from being overlooked, and practitioners could determine inspection focus areas, which could contribute to the improvement of maintenance techniques. In the future, it is expected to contribute to infrastructure diagnosis automation.

Learning the Meanings of Function Words From Grounded Language Using a Visual Question Answering Model.

Interpreting a seemingly simple function word like "or," "behind," or "more" can require logical, numerical, and relational reasoning. How are such words learned by children? Prior acquisition theories have often relied on positing a foundation of innate knowledge. Yet recent neural-network-based visual question answering models apparently can learn to use function words as part of answering questions about complex visual scenes. In this paper, we study what these models learn about function words, in the hope of better understanding how the meanings of these words can be learned by both models and children. We show that recurrent models trained on visually grounded language learn gradient semantics for function words requiring spatial and numerical reasoning. Furthermore, we find that these models can learn the meanings of logical connectives and and or without any prior knowledge of logical reasoning as well as early evidence that they are sensitive to alternative expressions when interpreting language. Finally, we show that word learning difficulty is dependent on the frequency of models' input. Our findings offer proof-of-concept evidence that it is possible to learn the nuanced interpretations of function words in a visually grounded context by using non-symbolic general statistical learning algorithms, without any prior knowledge of linguisticmeaning.

Visual Mind: Visual Question Answering (VQA) with CLIP Model

Abstract: This paper proposes a Visual Question Answering (VQA) problem using CLIP models. The proposed approach suggests an enhanced VQA-CLIP model with additional layers for better computational performance. VQA is an increasingly important task that aims to answer open-ended questions based on images. This task has numerous applications in various fields such as medicine, education, and surveillance. The VizWiz dataset, specifically designed to assist visually impaired individuals, consists of image/question pairs along with 10 answers per question, recorded by blind participants in a natural setting. The task involves predicting answers to questions and determining when a question is unanswerable. In this study, we will utilize the VizWiz dataset and employ the CLIP model with an additional linear layer, a multimodal, zero-shot model known for its efficiency in processing image and text data. By leveraging the unique capabilities of CLIP, and benchmarked against state-ofthe-art approaches. Results indicate a competitive or better performance of the VQA model.

Dual modality prompt learning for visual question-grounded answering in robotic surgery

With recent advancements in robotic surgery, notable strides have been made in visual question answering (VQA). Existing VQA systems typically generate textual answers to questions but fail to indicate the location of the relevant content within the image. This limitation restricts the interpretative capacity of the VQA models and their ability to explore specific image regions. To address this issue, this study proposes a grounded VQA model for robotic surgery, capable of localizing a specific region during answer prediction. Drawing inspiration from prompt learning in language models, a dual-modality prompt model was developed to enhance precise multimodal information interactions. Specifically, two complementary prompters were introduced to effectively integrate visual and textual prompts into the encoding process of the model. A visual complementary prompter merges visual prompt knowledge with visual information features to guide accurate localization. The textual complementary prompter aligns visual information with textual prompt knowledge and textual information, guiding textual information towards a more accurate inference of the answer. Additionally, a multiple iterative fusion strategy was adopted for comprehensive answer reasoning, to ensure high-quality generation of textual and grounded answers. The experimental results validate the effectiveness of the model, demonstrating its superiority over existing methods on the EndoVis-18 and EndoVis-17 datasets.

NExT-OOD: Overcoming Dual Multiple-Choice VQA Biases.

In recent years, multiple-choice Visual Question Answering (VQA) has become topical and achieved remarkable progress. However, most pioneer multiple-choice VQA models are heavily driven by statistical correlations in datasets, which cannot perform well on multimodal understanding and suffer from poor generalization. In this paper, we identify two kinds of spurious correlations, i.e., a Vision-Answer bias (VA bias) and a Question-Answer bias (QA bias). To systematically and scientifically study these biases, we construct a new video question answering (videoQA) benchmark NExT-OOD in OOD setting and propose a graph-based cross-sample method for bias reduction. Specifically, the NExT-OOD is designed to quantify models' generalizability and measure their reasoning ability comprehensively. It contains three sub-datasets including NExT-OOD-VA, NExT-OOD-QA, and NExT-OOD-VQA, which are designed for the VA bias, QA bias, and VA&QA bias, respectively. We evaluate several existing multiple-choice VQA models on our NExT-OOD, and illustrate that their performance degrades significantly compared with the results obtained on the original multiple-choice VQA dataset. Besides, to mitigate the VA bias and QA bias, we explicitly consider the cross-sample information and design a contrastive graph matching loss in our approach, which provides adequate debiasing guidance from the perspective of whole dataset, and encourages the model to focus on multimodal contents instead of spurious statistical regularities. Extensive experimental results illustrate that our method significantly outperforms other bias reduction strategies, demonstrating the effectiveness and generalizability of the proposed approach.

Improving Automatic VQA Evaluation Using Large Language Models

8 years after the visual question answering (VQA) task was proposed, accuracy remains the primary metric for automatic evaluation. VQA Accuracy has been effective so far in the IID evaluation setting. However, our community is undergoing a shift towards open-ended generative models and OOD evaluation. In this new paradigm, the existing VQA Accuracy metric is overly stringent and underestimates the performance of VQA systems. Thus, there is a need to develop more robust automatic VQA metrics that serve as a proxy for human judgment. In this work, we propose to leverage the in-context learning capabilities of instruction-tuned large language models (LLMs) to build a better VQA metric. We formulate VQA evaluation as an answer-rating task where the LLM is instructed to score the accuracy of a candidate answer given a set of reference answers. We demonstrate the proposed metric better correlates with human judgment compared to existing metrics across several VQA models and benchmarks. We hope wide adoption of our metric will contribute to better estimating the research progress on the VQA task. We plan to release the evaluation code and collected human judgments.