We analyze state-of-the-art deep learning models for three tasks: question answering on (1) images, (2) tables, and (3) passages of text. Using the notion of \emph{attribution} (word importance), we find that these deep networks often ignore important question terms. Leveraging such behavior, we perturb questions to craft a variety of adversarial examples. Our strongest attacks drop the accuracy of a visual question answering model from $61.1\%$ to $19\%$, and that of a tabular question answering model from $33.5\%$ to $3.3\%$. Additionally, we show how attributions can strengthen attacks proposed by Jia and Liang (2017) on paragraph comprehension models. Our results demonstrate that attributions can augment standard measures of accuracy and empower investigation of model performance. When a model is accurate but for the wrong reasons, attributions can surface erroneous logic in the model that indicates inadequacies in the test data.
In-depth scene descriptions and question answering tasks have greatly increased the scope of today's definition of scene understanding. While such tasks are in principle open ended, current formulations primarily focus on describing only the current state of the scenes under consideration. In contrast, in this paper, we focus on the future states of the scenes which are also conditioned on actions. We posit this as a question answering task, where an answer has to be given about a future scene state, given observations of the current scene, and a question that includes a hypothetical action. Our solution is a hybrid model which integrates a physics engine into a question answering architecture in order to anticipate future scene states resulting from object-object interactions caused by an action. We demonstrate first results on this challenging new problem and compare to baselines, where we outperform fully data-driven end-to-end learning approaches.
Medical visual question answering models can answer correctly without attending to the anatomy that the question names. Automatic region priors allow spatial supervision without expert masks, but they are noisy, and it is unclear how strongly they should shape attention and how their effect should be verified. We present QCAP-VQA, which supervises a subset of cross-attention heads with a question-conditioned prior built from ranked segmentation proposals. Its capped prior-mass objective penalizes insufficient attention inside the prior only below a threshold. We evaluate attention against SLAKE structure boxes that play no role in training, separating attention on the named anatomy from how attention shifts between questions about the same image. QCAP-VQA matches the accuracy of a baseline without spatial supervision (84.26% vs. 84.23%), and every evaluated prior-supervised model places more attention on the annotated structures. Over a 64-fold range of regularization strength, the capped objective keeps prior mass between 0.67 and 0.77, whereas a pooled region-mass objective drives it to 0.97 and loses accuracy. The cost lies in question conditioning: as the constraint widens or strengthens, attention shifts agree less with the expert-defined contrast. Across 13 configurations, prior-referenced and expert-referenced evaluation rank the models in largely opposite order (Spearman correlation of -0.87), and the capped models it favors track the prior's own errors. Because clinical use of medical visual question answering requires answers grounded in the queried anatomy, supervision from automatic priors should be judged against independent anatomical references.
The Document-based Visual Question Answering competition addresses the automatic detection of parent-child relationships between elements in multi-page documents. The goal is to identify the document elements that answer a specific question posed in natural language. This paper describes the PoliTo's approach to addressing this task, in particular, our best solution explores a text-only approach, leveraging an ad hoc sampling strategy. Specifically, our approach leverages the Masked Language Modeling technique to fine-tune a BERT model, focusing on sentences containing sensitive keywords that also occur in the questions, such as references to tables or images. Thanks to the effectiveness of this approach, we are able to achieve high performance compared to baselines, demonstrating how our solution contributes positively to this task.
Large video models have exhibited impressive performance on a wide range of visual question answering tasks, owing to the rise of powerful, pretrained text and vision encoders. The usefulness of such models have also been demonstrated on a wide range of benchmarks, with an important caveat - the dominant approach in these benchmarks evaluates multiple choice reasoning via text options. This is a natural way to test text-based reasoning in these models, and has led to significant insights regarding model behavior in the community. In this work, we ask a different question - what happens when the evaluation modality is visual, rather than text? We introduce three new vision-centric evaluation benchmarks in temporal frame retrieval, video future prediction, and causal memory distortion, all designed around evaluating visual understanding capabilities in large video models. Our approach complements the existing approaches to evaluate video understanding in frontier models. We show that current frontier models exhibit significant weakness when attempting to reason through visual queries, rather than text. We conclude with an extended analysis section that provides pointers for future improvements in visual understanding for large video models.
Can Visual Question Answering (VQA) systems perform just as well when deployed in the real world? Or are they susceptible to realistic corruption effects e.g. image blur, which can be detrimental in sensitive applications, such as medical VQA? While linguistic or textual robustness has been thoroughly explored in the VQA literature, there has yet to be any significant work on the visual robustness of VQA models. We propose the first large-scale benchmark comprising 213,000 augmented images, challenging the visual robustness of multiple VQA models and assessing the strength of realistic visual corruptions. Additionally, we have designed several robustness evaluation metrics that can be aggregated into a unified metric and tailored to fit a variety of use cases. Our experiments reveal several insights into the relationships between model size, performance, and robustness with the visual corruptions. Our benchmark highlights the need for a balanced approach in model development that considers model performance without compromising the robustness.
Understanding infographic charts with design-driven visual elements (e.g., pictograms, icons) requires both visual recognition and reasoning, posing challenges for multimodal large language models (MLLMs). However, existing visual-question answering benchmarks fall short in evaluating these capabilities of MLLMs due to the lack of paired plain charts and visual-element-based questions. To bridge this gap, we introduce InfoChartQA, a benchmark for evaluating MLLMs on infographic chart understanding. It includes 5,642 pairs of infographic and plain charts, each sharing the same underlying data but differing in visual presentations. We further design visual-element-based questions to capture their unique visual designs and communicative intent. Evaluation of 20 MLLMs reveals a substantial performance decline on infographic charts, particularly for visual-element-based questions related to metaphors. The paired infographic and plain charts enable fine-grained error analysis and ablation studies, which highlight new opportunities for advancing MLLMs in infographic chart understanding. We release InfoChartQA at https://github.com/CoolDawnAnt/InfoChartQA.
Generalization to out-of-distribution data has been a problem for Visual Question Answering (VQA) models. To measure generalization to novel questions, we propose to separate them into "skills" and "concepts". "Skills" are visual tasks, such as counting or attribute recognition, and are applied to "concepts" mentioned in the question, such as objects and people. VQA methods should be able to compose skills and concepts in novel ways, regardless of whether the specific composition has been seen in training, yet we demonstrate that existing models have much to improve upon towards handling new compositions. We present a novel method for learning to compose skills and concepts that separates these two factors implicitly within a model by learning grounded concept representations and disentangling the encoding of skills from that of concepts. We enforce these properties with a novel contrastive learning procedure that does not rely on external annotations and can be learned from unlabeled image-question pairs. Experiments demonstrate the effectiveness of our approach for improving compositional and grounding performance.
We introduce the task of Image-Set Visual Question Answering (ISVQA), which generalizes the commonly studied single-image VQA problem to multi-image settings. Taking a natural language question and a set of images as input, it aims to answer the question based on the content of the images. The questions can be about objects and relationships in one or more images or about the entire scene depicted by the image set. To enable research in this new topic, we introduce two ISVQA datasets - indoor and outdoor scenes. They simulate the real-world scenarios of indoor image collections and multiple car-mounted cameras, respectively. The indoor-scene dataset contains 91,479 human annotated questions for 48,138 image sets, and the outdoor-scene dataset has 49,617 questions for 12,746 image sets. We analyze the properties of the two datasets, including question-and-answer distributions, types of questions, biases in dataset, and question-image dependencies. We also build new baseline models to investigate new research challenges in ISVQA.
Explainable artificial intelligence (XAI) aims to make machine learning models more transparent. While many approaches focus on generating explanations post-hoc, interpretable approaches, which generate the explanations intrinsically alongside the predictions, are relatively rare. In this work, we integrate different discrete subset sampling methods into a graph-based visual question answering system to compare their effectiveness in generating interpretable explanatory subgraphs intrinsically. We evaluate the methods on the GQA dataset and show that the integrated methods effectively mitigate the performance trade-off between interpretability and answer accuracy, while also achieving strong co-occurrences between answer and question tokens. Furthermore, we conduct a human evaluation to assess the interpretability of the generated subgraphs using a comparative setting with the extended Bradley-Terry model, showing that the answer and question token co-occurrence metrics strongly correlate with human preferences. Our source code is publicly available.
We present LinkQ, a system that leverages a large language model (LLM) to facilitate knowledge graph (KG) query construction through natural language question-answering. Traditional approaches often require detailed knowledge of complex graph querying languages, limiting the ability for users -- even experts -- to acquire valuable insights from KG data. LinkQ simplifies this process by first interpreting a user's question, then converting it into a well-formed KG query. By using the LLM to construct a query instead of directly answering the user's question, LinkQ guards against the LLM hallucinating or generating false, erroneous information. By integrating an LLM into LinkQ, users are able to conduct both exploratory and confirmatory data analysis, with the LLM helping to iteratively refine open-ended questions into precise ones. To demonstrate the efficacy of LinkQ, we conducted a qualitative study with five KG practitioners and distill their feedback. Our results indicate that practitioners find LinkQ effective for KG question-answering, and desire future LLM-assisted systems for the exploratory analysis of graph databases.
Visual question answering (VQA) is known as an AI-complete task as it requires understanding, reasoning, and inferring about the vision and the language content. Over the past few years, numerous neural architectures have been suggested for the VQA problem. However, achieving success in zero-shot VQA remains a challenge due to its requirement for advanced generalization and reasoning skills. This study explores the impact of incorporating image captioning as an intermediary process within the VQA pipeline. Specifically, we explore the efficacy of utilizing image captions instead of images and leveraging large language models (LLMs) to establish a zero-shot setting. Since image captioning is the most crucial step in this process, we compare the impact of state-of-the-art image captioning models on VQA performance across various question types in terms of structure and semantics. We propose a straightforward and efficient question-driven image captioning approach within this pipeline to transfer contextual information into the question-answering (QA) model. This method involves extracting keywords from the question, generating a caption for each image-question pair using the keywords, and incorporating the question-driven caption into the LLM prompt. We evaluate the efficacy of using general-purpose and question-driven image captions in the VQA pipeline. Our study highlights the potential of employing image captions and harnessing the capabilities of LLMs to achieve competitive performance on GQA under the zero-shot setting. Our code is available at \url{https://github.com/ovguyo/captions-in-VQA}.
Vision-language models, while effective in general domains and showing strong performance in diverse multi-modal applications like visual question-answering (VQA), struggle to maintain the same level of effectiveness in more specialized domains, e.g., medical. We propose a medical vision-language model that integrates large vision and language models adapted for the medical domain. This model goes through three stages of parameter-efficient training using three separate biomedical and radiology multi-modal visual and text datasets. The proposed model achieves state-of-the-art performance on the SLAKE 1.0 medical VQA (MedVQA) dataset with an overall accuracy of 87.5% and demonstrates strong performance on another MedVQA dataset, VQA-RAD, achieving an overall accuracy of 73.2%.
Multimodal Large Language Models (MLLMs) have experienced rapid progress in visual recognition tasks in recent years. Given their potential integration into many critical applications, it is important to understand the limitations of their visual perception. In this work, we study whether MLLMs can perceive small visual details as effectively as large ones when answering questions about images. We observe that their performance is very sensitive to the size of the visual subject of the question, and further show that this effect is in fact causal by conducting an intervention study. Next, we study the attention patterns of MLLMs when answering visual questions, and intriguingly find that they consistently know where to look, even when they provide the wrong answer. Based on these findings, we then propose training-free visual intervention methods that leverage the internal knowledge of any MLLM itself, in the form of attention and gradient maps, to enhance its perception of small visual details. We evaluate our proposed methods on two widely-used MLLMs and seven visual question answering benchmarks and show that they can significantly improve MLLMs' accuracy without requiring any training. Our results elucidate the risk of applying MLLMs to visual recognition tasks concerning small details and indicate that visual intervention using the model's internal state is a promising direction to mitigate this risk.
Visual Question Answering (VQA) models play a critical role in enhancing the perception capabilities of autonomous driving systems by allowing vehicles to analyze visual inputs alongside textual queries, fostering natural interaction and trust between the vehicle and its occupants or other road users. This study investigates the attention patterns of humans compared to a VQA model when answering driving-related questions, revealing disparities in the objects observed. We propose an approach integrating filters to optimize the model's attention mechanisms, prioritizing relevant objects and improving accuracy. Utilizing the LXMERT model for a case study, we compare attention patterns of the pre-trained and Filter Integrated models, alongside human answers using images from the NuImages dataset, gaining insights into feature prioritization. We evaluated the models using a Subjective scoring framework which shows that the integration of the feature encoder filter has enhanced the performance of the VQA model by refining its attention mechanisms.
Multimodal large language models (MLLMs) have demonstrated great performance on visual question answering (VQA). When it comes to knowledge-based Visual Question Answering (KB-VQA), MLLMs may lack the specialized domain knowledge needed to answer questions, necessitating the retrieval of necessary information from external knowledge sources. Previous works like Retrival-Augmented VQA-v2 (RAVQA-v2) focus on utilizing as much input information, such as image-based textual descriptions and retrieved knowledge, as possible to improve performance, but they all overlook the issue that with the number of input tokens increasing, inference efficiency significantly decreases, which contradicts the demands of practical applications. To address this issue, we propose \textbf{R}etrieval-\textbf{A}ugmented MLLMs with Compressed Contexts (RACC). RACC learns to compress and aggregate retrieved knowledge for a given image-question pair, generating a compact modulation in the form of Key-Value (KV) cache to adapt the downstream frozen MLLM, thereby achieving effective and efficient inference. RACC achieves a state-of-the-art (SOTA) performance of 63.92\% on OK-VQA. Moreover, it significantly reduces inference latency by 22.0\%-59.7\% compared to the prominent RAVQA-v2. Abundant experiments show RACC's broad applicability. It is compatible with various off-the-shelf MLLMs and can also handle different knowledge sources including textual and multimodal documents.
Knowledge-based visual question answering (VQA) requires answering questions with external knowledge in addition to the content of images. One dataset that is mostly used in evaluating knowledge-based VQA is OK-VQA, but it lacks a gold standard knowledge corpus for retrieval. Existing work leverage different knowledge bases (e.g., ConceptNet and Wikipedia) to obtain external knowledge. Because of varying knowledge bases, it is hard to fairly compare models' performance. To address this issue, we collect a natural language knowledge base that can be used for any VQA system. Moreover, we propose a Visual Retriever-Reader pipeline to approach knowledge-based VQA. The visual retriever aims to retrieve relevant knowledge, and the visual reader seeks to predict answers based on given knowledge. We introduce various ways to retrieve knowledge using text and images and two reader styles: classification and extraction. Both the retriever and reader are trained with weak supervision. Our experimental results show that a good retriever can significantly improve the reader's performance on the OK-VQA challenge. The code and corpus are provided in https://github.com/luomancs/retriever\_reader\_for\_okvqa.git
CLIP has shown a remarkable zero-shot capability on a wide range of vision tasks. Previously, CLIP is only regarded as a powerful visual encoder. However, after being pre-trained by language supervision from a large amount of image-caption pairs, CLIP itself should also have acquired some few-shot abilities for vision-language tasks. In this work, we empirically show that CLIP can be a strong vision-language few-shot learner by leveraging the power of language. We first evaluate CLIP's zero-shot performance on a typical visual question answering task and demonstrate a zero-shot cross-modality transfer capability of CLIP on the visual entailment task. Then we propose a parameter-efficient fine-tuning strategy to boost the few-shot performance on the vqa task. We achieve competitive zero/few-shot results on the visual question answering and visual entailment tasks without introducing any additional pre-training procedure.
We present an effective method for fusing visual-and-language representations for several question answering tasks including visual question answering and visual entailment. In contrast to prior works that concatenate unimodal representations or use only cross-attention, we compose multimodal representations via channel fusion. By fusing on the channels, the model is able to more effectively align the tokens compared to standard methods. These multimodal representations, which we call compound tokens are generated with cross-attention transformer layers. First, vision tokens are used as queries to retrieve compatible text tokens through cross-attention. We then chain the vision tokens and the queried text tokens along the channel dimension. We call the resulting representations compound tokens. A second group of compound tokens are generated using an analogous process where the text tokens serve as queries to the cross-attention layer. We concatenate all the compound tokens for further processing with multimodal encoder. We demonstrate the effectiveness of compound tokens using an encoder-decoder vision-language model trained end-to-end in the open-vocabulary setting. Compound Tokens achieve highly competitive performance across a range of question answering tasks including GQA, VQA2.0, and SNLI-VE.
Visual Question Answering (VQA) research is split into two camps: the first focuses on VQA datasets that require natural image understanding and the second focuses on synthetic datasets that test reasoning. A good VQA algorithm should be capable of both, but only a few VQA algorithms are tested in this manner. We compare five state-of-the-art VQA algorithms across eight VQA datasets covering both domains. To make the comparison fair, all of the models are standardized as much as possible, e.g., they use the same visual features, answer vocabularies, etc. We find that methods do not generalize across the two domains. To address this problem, we propose a new VQA algorithm that rivals or exceeds the state-of-the-art for both domains.