Noa Garcia

Yusuke Hirota, Yuta Nakashima, Noa Garcia

Image captioning models are known to perpetuate and amplify harmful societal bias in the training set. In this work, we aim to mitigate such gender bias in image captioning models. While prior work has addressed this problem by forcing models to focus on people to reduce gender misclassification, it conversely generates gender-stereotypical words at the expense of predicting the correct gender. From this observation, we hypothesize that there are two types of gender bias affecting image captioning models: 1) bias that exploits context to predict gender, and 2) bias in the probability of generating certain (often stereotypical) words because of gender. To mitigate both types of gender biases, we propose a framework, called LIBRA, that learns from synthetically biased samples to decrease both types of biases, correcting gender misclassification and changing gender-stereotypical words to more neutral ones. Code is available at https://github.com/rebnej/LIBRA.

Piera Riccio, Miriam Doh, Benedikt Höltgen et al.

Artistic styles are rooted in specific socio-historical contexts that encode social hierarchies, including distinct constructions of gender. Yet in AI research, style has long been treated as a surface-level visual property: a filter of color, brushstroke, and texture applied to otherwise content-neutral scenes. We introduce the first dataset to investigate the interplay between gender representation and style in both historical and generated images. StyleGender comprises 74k images spanning 19 artistic styles, comprising art historical images with style and gender annotations, T2I-generated images under controlled style and gender prompts, and a semantically aligned set enabling direct art history-to-generation comparison. By proposing two Set Gender Artifact (SGA) metrics (PixelSGA and MaskSGA), capturing gender signals at the pixel level and in compositional structure, we show that (1) gender representation shapes visual features across artistic styles, (2) style keywords carry these patterns into T2I generation, and (3) generative models tend to amplify gender artifacts beyond what is observed in historical sources.

Noa Garcia, Yusuke Hirota, Yankun Wu et al.

The increasing tendency to collect large and uncurated datasets to train vision-and-language models has raised concerns about fair representations. It is known that even small but manually annotated datasets, such as MSCOCO, are affected by societal bias. This problem, far from being solved, may be getting worse with data crawled from the Internet without much control. In addition, the lack of tools to analyze societal bias in big collections of images makes addressing the problem extremely challenging. Our first contribution is to annotate part of the Google Conceptual Captions dataset, widely used for training vision-and-language models, with four demographic and two contextual attributes. Our second contribution is to conduct a comprehensive analysis of the annotations, focusing on how different demographic groups are represented. Our last contribution lies in evaluating three prevailing vision-and-language tasks: image captioning, text-image CLIP embeddings, and text-to-image generation, showing that societal bias is a persistent problem in all of them.

Yusuke Hirota, Yuta Nakashima, Noa Garcia

Vision-and-language tasks have increasingly drawn more attention as a means to evaluate human-like reasoning in machine learning models. A popular task in the field is visual question answering (VQA), which aims to answer questions about images. However, VQA models have been shown to exploit language bias by learning the statistical correlations between questions and answers without looking into the image content: e.g., questions about the color of a banana are answered with yellow, even if the banana in the image is green. If societal bias (e.g., sexism, racism, ableism, etc.) is present in the training data, this problem may be causing VQA models to learn harmful stereotypes. For this reason, we investigate gender and racial bias in five VQA datasets. In our analysis, we find that the distribution of answers is highly different between questions about women and men, as well as the existence of detrimental gender-stereotypical samples. Likewise, we identify that specific race-related attributes are underrepresented, whereas potentially discriminatory samples appear in the analyzed datasets. Our findings suggest that there are dangers associated to using VQA datasets without considering and dealing with the potentially harmful stereotypes. We conclude the paper by proposing solutions to alleviate the problem before, during, and after the dataset collection process.

Yusuke Hirota, Yuta Nakashima, Noa Garcia

We study societal bias amplification in image captioning. Image captioning models have been shown to perpetuate gender and racial biases, however, metrics to measure, quantify, and evaluate the societal bias in captions are not yet standardized. We provide a comprehensive study on the strengths and limitations of each metric, and propose LIC, a metric to study captioning bias amplification. We argue that, for image captioning, it is not enough to focus on the correct prediction of the protected attribute, and the whole context should be taken into account. We conduct extensive evaluation on traditional and state-of-the-art image captioning models, and surprisingly find that, by only focusing on the protected attribute prediction, bias mitigation models are unexpectedly amplifying bias.

Yankun Wu, Yuta Nakashima, Noa Garcia

The duality of content and style is inherent to the nature of art. For humans, these two elements are clearly different: content refers to the objects and concepts in the piece of art, and style to the way it is expressed. This duality poses an important challenge for computer vision. The visual appearance of objects and concepts is modulated by the style that may reflect the author's emotions, social trends, artistic movement, etc., and their deep comprehension undoubtfully requires to handle both. A promising step towards a general paradigm for art analysis is to disentangle content and style, whereas relying on human annotations to cull a single aspect of artworks has limitations in learning semantic concepts and the visual appearance of paintings. We thus present GOYA, a method that distills the artistic knowledge captured in a recent generative model to disentangle content and style. Experiments show that synthetically generated images sufficiently serve as a proxy of the real distribution of artworks, allowing GOYA to separately represent the two elements of art while keeping more information than existing methods.

Tong Xiang, Liangzhi Li, Wangyue Li et al.

The recent advances in natural language processing (NLP), have led to a new trend of applying large language models (LLMs) to real-world scenarios. While the latest LLMs are astonishingly fluent when interacting with humans, they suffer from the misinformation problem by unintentionally generating factually false statements. This can lead to harmful consequences, especially when produced within sensitive contexts, such as healthcare. Yet few previous works have focused on evaluating misinformation in the long-form (LF) generation of LLMs, especially for knowledge-intensive topics. Moreover, although LLMs have been shown to perform well in different languages, misinformation evaluation has been mostly conducted in English. To this end, we present a benchmark, CARE-MI, for evaluating LLM misinformation in: 1) a sensitive topic, specifically the maternity and infant care domain; and 2) a language other than English, namely Chinese. Most importantly, we provide an innovative paradigm for building LF generation evaluation benchmarks that can be transferred to other knowledge-intensive domains and low-resourced languages. Our proposed benchmark fills the gap between the extensive usage of LLMs and the lack of datasets for assessing the misinformation generated by these models. It contains 1,612 expert-checked questions, accompanied with human-selected references. Using our benchmark, we conduct extensive experiments and found that current Chinese LLMs are far from perfect in the topic of maternity and infant care. In an effort to minimize the reliance on human resources for performance evaluation, we offer off-the-shelf judgment models for automatically assessing the LF output of LLMs given benchmark questions. Moreover, we compare potential solutions for LF generation evaluation and provide insights for building better automated metrics.

Tianwei Chen, Noa Garcia, Mayu Otani et al.

Is more data always better to train vision-and-language models? We study knowledge transferability in multi-modal tasks. The current tendency in machine learning is to assume that by joining multiple datasets from different tasks their overall performance will improve. However, we show that not all the knowledge transfers well or has a positive impact on related tasks, even when they share a common goal. We conduct an exhaustive analysis based on hundreds of cross-experiments on 12 vision-and-language tasks categorized in 4 groups. Whereas tasks in the same group are prone to improve each other, results show that this is not always the case. Other factors such as dataset size or pre-training stage have also a great impact on how well the knowledge is transferred.

Wangyue Li, Liangzhi Li, Tong Xiang et al.

Multiple-choice questions (MCQs) are widely used in the evaluation of large language models (LLMs) due to their simplicity and efficiency. However, there are concerns about whether MCQs can truly measure LLM's capabilities, particularly in knowledge-intensive scenarios where long-form generation (LFG) answers are required. The misalignment between the task and the evaluation method demands a thoughtful analysis of MCQ's efficacy, which we undertake in this paper by evaluating nine LLMs on four question-answering (QA) datasets in two languages: Chinese and English. We identify a significant issue: LLMs exhibit an order sensitivity in bilingual MCQs, favoring answers located at specific positions, i.e., the first position. We further quantify the gap between MCQs and long-form generation questions (LFGQs) by comparing their direct outputs, token logits, and embeddings. Our results reveal a relatively low correlation between answers from MCQs and LFGQs for identical questions. Additionally, we propose two methods to quantify the consistency and confidence of LLMs' output, which can be generalized to other QA evaluation benchmarks. Notably, our analysis challenges the idea that the higher the consistency, the greater the accuracy. We also find MCQs to be less reliable than LFGQs in terms of expected calibration error. Finally, the misalignment between MCQs and LFGQs is not only reflected in the evaluation performance but also in the embedding space. Our code and models can be accessed at https://github.com/Meetyou-AI-Lab/Can-MC-Evaluate-LLMs.

Lu Wei, Yuta Nakashima, Noa Garcia

The widespread adoption of text-to-image (T2I) generation has raised concerns about privacy, bias, and copyright violations. Concept erasure techniques offer a promising solution by selectively removing undesired concepts from pre-trained models without requiring full retraining. However, these methods are often evaluated on a limited set of concepts, relying on overly simplistic and direct prompts. To test the boundaries of concept erasure techniques, and assess whether they truly remove targeted concepts from model representations, we introduce EMMA, a benchmark that evaluates five key dimensions of concept erasure over 12 metrics. EMMA goes beyond standard metrics like image quality and time efficiency, testing robustness under challenging conditions, including indirect descriptions, visually similar non-target concepts, and potential gender and ethnicity bias, providing a socially aware analysis of method behavior. Using EMMA, we analyze five concept erasure methods across five domains (objects, celebrities, art styles, NSFW, and copyright). Our results show that existing methods struggle with implicit prompts (i.e., generating the erased concept when it is indirectly referenced) and visually similar non-target concepts (i.e., failing to generate non-targeted concepts resembling the erased one), while some amplify gender and ethnicity bias compared to the original model.

Xiao Liu, Liangzhi Li, Tong Xiang et al.

With the development of large language models (LLMs) like ChatGPT, both their vast applications and potential vulnerabilities have come to the forefront. While developers have integrated multiple safety mechanisms to mitigate their misuse, a risk remains, particularly when models encounter adversarial inputs. This study unveils an attack mechanism that capitalizes on human conversation strategies to extract harmful information from LLMs. We delineate three pivotal strategies: (i) decomposing malicious questions into seemingly innocent sub-questions; (ii) rewriting overtly malicious questions into more covert, benign-sounding ones; (iii) enhancing the harmfulness of responses by prompting models for illustrative examples. Unlike conventional methods that target explicit malicious responses, our approach delves deeper into the nature of the information provided in responses. Through our experiments conducted on GPT-3.5-turbo, GPT-4, and Llama2, our method has demonstrated a marked efficacy compared to conventional attack methods. In summary, this work introduces a novel attack method that outperforms previous approaches, raising an important question: How to discern whether the ultimate intent in a dialogue is malicious?

Jose Luna, Yankun Wu, Xiaofei Xie et al.

Text-to-image (T2I) generative models are increasingly used to produce content for education, media, and public-facing communication, and are starting to be integrated into higher-impact pipelines. Since generated images tend to reinforce stereotypes, producing representational erasure via "default" depictions and shaping perceptions of who belongs in certain roles, a growing body of work has proposed metrics to quantify gender bias in T2I outputs. Yet existing evaluations remain fragmented. Metrics are often reported without a shared view of what they measure, what assumptions they entail, or how their results should be interpreted under different deployment contexts. This limits the usefulness of gender bias measurement for both technical auditing and emerging governance discussions. We propose a risk-aligned auditing framework for gender bias in T2I models composed of three constituents that connects risk categories, evaluation metrics, and harms. First, we identify risk-tiered use-case profiles aligned with the EU AI Act's risk categories to motivate why auditing expectations may vary with deployment contexts and stakeholder exposure. Second, we construct a metric catalog that consolidates gender-bias evaluation methods and organizes them in three measurement categories: gender prediction, embedding similarity, and downstream task. Third, we introduce a harm typology that maps context-dependent harm categories (e.g., representational, quality-of-service) to specific risk-tired scenarios. Finally, we introduce THUMB cards (Text-to-image Harms-informed Use-case-aligned Metrics of gender Bias) that help formulate auditing systematically by the incorporation of context, scenario and bias manifestation, harm hypotheses, and audit strategy.

Rawisara Lohanimit, Yankun Wu, Amelia Katirai et al.

The rise of generative models has led to increased use of large-scale datasets collected from the internet, often with minimal or no data curation. This raises concerns about the inclusion of sensitive or private information. In this work, we explore the presence of pregnancy ultrasound images, which contain sensitive personal information and are often shared online. Through a systematic examination of LAION-400M dataset using CLIP embedding similarity, we retrieve images containing pregnancy ultrasound and detect thousands of entities of private information such as names and locations. Our findings reveal that multiple images have high-risk information that could enable re-identification or impersonation. We conclude with recommended practices for dataset curation, data privacy, and ethical use of public image datasets.

Ryan Ramos, Vladan Stojnić, Giorgos Kordopatis-Zilos et al.

Prior work has analyzed the robustness of visual encoders to image transformations and corruptions, particularly in cases where such alterations are not seen during training. When this occurs, they introduce a form of distribution shift at test time, often leading to performance degradation. The primary focus has been on severe corruptions that, when applied aggressively, distort useful signals necessary for accurate semantic predictions. We take a different perspective by analyzing parameters of the image acquisition process and transformations that may be subtle or even imperceptible to the human eye. We find that such parameters are systematically encoded in the learned visual representations and can be easily recovered. More strikingly, their presence can have a profound impact, either positively or negatively, on semantic predictions. This effect depends on whether there is a strong correlation or anti-correlation between semantic labels and these acquisition-based or processing-based labels. Our code and data are available at: https://github.com/ryan-caesar-ramos/visual-encoder-traces

Patrick Ramos, Nicolas Gonthier, Selina Khan et al.

Object detection in art is a valuable tool for the digital humanities, as it allows for faster identification of objects in artistic and historical images compared to humans. However, annotating such images poses significant challenges due to the need for specialized domain expertise. We present NADA (no annotations for detection in art), a pipeline that leverages diffusion models' art-related knowledge for object detection in paintings without the need for full bounding box supervision. Our method, which supports both weakly-supervised and zero-shot scenarios and does not require any fine-tuning of its pretrained components, consists of a class proposer based on large vision-language models and a class-conditioned detector based on Stable Diffusion. NADA is evaluated on two artwork datasets, ArtDL 2.0 and IconArt, outperforming prior work in weakly-supervised detection, while being the first work for zero-shot object detection in art. Code is available at https://github.com/patrick-john-ramos/nada

Noa Garcia, Amelia Katirai

Computer vision, a core domain of artificial intelligence (AI), is the field that enables the computational analysis, understanding, and generation of visual data. Despite being historically rooted in military funding and increasingly deployed in warfare, the field tends to position itself as a neutral, purely technical endeavor, failing to engage in discussions about its dual-use applications. Yet it has been reported that computer vision systems are being systematically weaponized to assist in technologies that inflict harm, such as surveillance or warfare. Expanding on these concerns, we study the extent to which computer vision research is being used in the military and surveillance domains. We do so by collecting a dataset of tech companies with financial ties to the field's central research exchange platform: conferences. Conference sponsorship, we argue, not only serves as strong evidence of a company's investment in the field but also provides a privileged position for shaping its trajectory. By investigating sponsors' activities, we reveal that 44% of them have a direct connection with military or surveillance applications. We extend our analysis through two case studies in which we discuss the opportunities and limitations of sponsorship as a means for uncovering technological weaponization.

Yankun Wu, Yuta Nakashima, Noa Garcia

Several studies have raised awareness about social biases in image generative models, demonstrating their predisposition towards stereotypes and imbalances. This paper contributes to this growing body of research by introducing an evaluation protocol that analyzes the impact of gender indicators at every step of the generation process on Stable Diffusion images. Leveraging insights from prior work, we explore how gender indicators not only affect gender presentation but also the representation of objects and layouts within the generated images. Our findings include the existence of differences in the depiction of objects, such as instruments tailored for specific genders, and shifts in overall layouts. We also reveal that neutral prompts tend to produce images more aligned with masculine prompts than their feminine counterparts. We further explore where bias originates through representational disparities and how it manifests in the images via prompt-image dependencies, and provide recommendations for developers and users to mitigate potential bias in image generation.

Tianwei Chen, Yusuke Hirota, Mayu Otani et al.

We investigate the impact of deep generative models on potential social biases in upcoming computer vision models. As the internet witnesses an increasing influx of AI-generated images, concerns arise regarding inherent biases that may accompany them, potentially leading to the dissemination of harmful content. This paper explores whether a detrimental feedback loop, resulting in bias amplification, would occur if generated images were used as the training data for future models. We conduct simulations by progressively substituting original images in COCO and CC3M datasets with images generated through Stable Diffusion. The modified datasets are used to train OpenCLIP and image captioning models, which we evaluate in terms of quality and bias. Contrary to expectations, our findings indicate that introducing generated images during training does not uniformly amplify bias. Instead, instances of bias mitigation across specific tasks are observed. We further explore the factors that may influence these phenomena, such as artifacts in image generation (e.g., blurry faces) or pre-existing biases in the original datasets.

Piera Riccio, Francesco Galati, Kajetan Schweighofer et al.

In the era of large-scale visual data, understanding collections of images is a challenging yet important task. To this end, we introduce ImageSet2Text, a novel method to automatically generate natural language descriptions of image sets. Based on large language models, visual-question answering chains, an external lexical graph, and CLIP-based verification, ImageSet2Text iteratively extracts key concepts from image subsets and organizes them into a structured concept graph. We conduct extensive experiments evaluating the quality of the generated descriptions in terms of accuracy, completeness, and user satisfaction. We also examine the method's behavior through ablation studies, scalability assessments, and failure analyses. Results demonstrate that ImageSet2Text combines data-driven AI and symbolic representations to reliably summarize large image collections for a wide range of applications.

Yankun Wu, Zakaria Laskar, Giorgos Kordopatis-Zilos et al.

Instance-level recognition (ILR) focuses on identifying individual objects rather than broad categories, offering the highest granularity in image classification. However, this fine-grained nature makes creating large-scale annotated datasets challenging, limiting ILR's real-world applicability across domains. To overcome this, we introduce a novel approach that synthetically generates diverse object instances from multiple domains under varied conditions and backgrounds, forming a large-scale training set. Unlike prior work on automatic data synthesis, our method is the first to address ILR-specific challenges without relying on any real images. Fine-tuning foundation vision models on the generated data significantly improves retrieval performance across seven ILR benchmarks spanning multiple domains. Our approach offers a new, efficient, and effective alternative to extensive data collection and curation, introducing a new ILR paradigm where the only input is the names of the target domains, unlocking a wide range of real-world applications.

Yusuke Hirota, Ryo Hachiuma, Boyi Li et al. · uw

Gender bias in vision-language foundation models (VLMs) raises concerns about their safe deployment and is typically evaluated using benchmarks with gender annotations on real-world images. However, as these benchmarks often contain spurious correlations between gender and non-gender features, such as objects and backgrounds, we identify a critical oversight in gender bias evaluation: Do spurious features distort gender bias evaluation? To address this question, we systematically perturb non-gender features across four widely used benchmarks (COCO-gender, FACET, MIAP, and PHASE) and various VLMs to quantify their impact on bias evaluation. Our findings reveal that even minimal perturbations, such as masking just 10% of objects or weakly blurring backgrounds, can dramatically alter bias scores, shifting metrics by up to 175% in generative VLMs and 43% in CLIP variants. This suggests that current bias evaluations often reflect model responses to spurious features rather than gender bias, undermining their reliability. Since creating spurious feature-free benchmarks is fundamentally challenging, we recommend reporting bias metrics alongside feature-sensitivity measurements to enable a more reliable bias assessment.

Ryan Ramos, Yusuke Hirota, Yuta Nakashima et al.

The recycling of contrastive language-image pre-trained (CLIP) models as backbones for a large number of downstream tasks calls for a thorough analysis of their transferability implications, especially their well-documented reproduction of social biases and human stereotypes. How do such biases, learned during pre-training, propagate to downstream applications like visual question answering or image captioning? Do they transfer at all? We investigate this phenomenon, referred to as bias transfer in prior literature, through a comprehensive empirical analysis. Firstly, we examine how pre-training bias varies between global and local views of data, finding that bias measurement is highly dependent on the subset of data on which it is computed. Secondly, we analyze correlations between biases in the pre-trained models and the downstream tasks across varying levels of pre-training bias, finding difficulty in discovering consistent trends in bias transfer. Finally, we explore why this inconsistency occurs, showing that under the current paradigm, representation spaces of different pre-trained CLIPs tend to converge when adapted for downstream tasks. We hope this work offers valuable insights into bias behavior and informs future research to promote better bias mitigation practices.

Patrick Ramos, Ryan Ramos, Noa Garcia

We analyze data leakage in visual datasets. Data leakage refers to images in evaluation benchmarks that have been seen during training, compromising fair model evaluation. Given that large-scale datasets are often sourced from the internet, where many computer vision benchmarks are publicly available, our efforts are focused into identifying and studying this phenomenon. We characterize visual leakage into different types according to its modality, coverage, and degree. By applying image retrieval techniques, we unequivocally show that all the analyzed datasets present some form of leakage, and that all types of leakage, from severe instances to more subtle cases, compromise the reliability of model evaluation in downstream tasks.

Lu Wei, Liangzhi Li, Tong Xiang et al.

The internet has become a hotspot for hate speech (HS), threatening societal harmony and individual well-being. While automatic detection methods perform well in identifying explicit hate speech (ex-HS), they struggle with more subtle forms, such as implicit hate speech (im-HS). We tackle this problem by introducing a new taxonomy for im-HS detection, defining six encoding strategies named codetypes. We present two methods for integrating codetypes into im-HS detection: 1) prompting large language models (LLMs) directly to classify sentences based on generated responses, and 2) using LLMs as encoders with codetypes embedded during the encoding process. Experiments show that the use of codetypes improves im-HS detection in both Chinese and English datasets, validating the effectiveness of our approach across different languages.

Nikolaos-Antonios Ypsilantis, Noa Garcia, Guangxing Han et al.

This work introduces a dataset for large-scale instance-level recognition in the domain of artworks. The proposed benchmark exhibits a number of different challenges such as large inter-class similarity, long tail distribution, and many classes. We rely on the open access collection of The Met museum to form a large training set of about 224k classes, where each class corresponds to a museum exhibit with photos taken under studio conditions. Testing is primarily performed on photos taken by museum guests depicting exhibits, which introduces a distribution shift between training and testing. Testing is additionally performed on a set of images not related to Met exhibits making the task resemble an out-of-distribution detection problem. The proposed benchmark follows the paradigm of other recent datasets for instance-level recognition on different domains to encourage research on domain independent approaches. A number of suitable approaches are evaluated to offer a testbed for future comparisons. Self-supervised and supervised contrastive learning are effectively combined to train the backbone which is used for non-parametric classification that is shown as a promising direction. Dataset webpage: http://cmp.felk.cvut.cz/met/

Tianran Wu, Noa Garcia, Mayu Otani et al.

Video question answering (VideoQA) is designed to answer a given question based on a relevant video clip. The current available large-scale datasets have made it possible to formulate VideoQA as the joint understanding of visual and language information. However, this training procedure is costly and still less competent with human performance. In this paper, we investigate a transfer learning method by the introduction of domain-agnostic knowledge and domain-specific knowledge. First, we develop a novel transfer learning framework, which finetunes the pre-trained model by applying domain-agnostic knowledge as the medium. Second, we construct a new VideoQA dataset with 21,412 human-generated question-answer samples for comparable transfer of knowledge. Our experiments show that: (i) domain-agnostic knowledge is transferable and (ii) our proposed transfer learning framework can boost VideoQA performance effectively.

Zechen Bai, Yuta Nakashima, Noa Garcia

Have you ever looked at a painting and wondered what is the story behind it? This work presents a framework to bring art closer to people by generating comprehensive descriptions of fine-art paintings. Generating informative descriptions for artworks, however, is extremely challenging, as it requires to 1) describe multiple aspects of the image such as its style, content, or composition, and 2) provide background and contextual knowledge about the artist, their influences, or the historical period. To address these challenges, we introduce a multi-topic and knowledgeable art description framework, which modules the generated sentences according to three artistic topics and, additionally, enhances each description with external knowledge. The framework is validated through an exhaustive analysis, both quantitative and qualitative, as well as a comparative human evaluation, demonstrating outstanding results in terms of both topic diversity and information veracity.

Yusuke Hirota, Noa Garcia, Mayu Otani et al.

How far can we go with textual representations for understanding pictures? In image understanding, it is essential to use concise but detailed image representations. Deep visual features extracted by vision models, such as Faster R-CNN, are prevailing used in multiple tasks, and especially in visual question answering (VQA). However, conventional deep visual features may struggle to convey all the details in an image as we humans do. Meanwhile, with recent language models' progress, descriptive text may be an alternative to this problem. This paper delves into the effectiveness of textual representations for image understanding in the specific context of VQA. We propose to take description-question pairs as input, instead of deep visual features, and fed them into a language-only Transformer model, simplifying the process and the computational cost. We also experiment with data augmentation techniques to increase the diversity in the training set and avoid learning statistical bias. Extensive evaluations have shown that textual representations require only about a hundred words to compete with deep visual features on both VQA 2.0 and VQA-CP v2.

Cheikh Brahim El Vaigh, Noa Garcia, Benjamin Renoust et al.

The rise of digitization of cultural documents offers large-scale contents, opening the road for development of AI systems in order to preserve, search, and deliver cultural heritage. To organize such cultural content also means to classify them, a task that is very familiar to modern computer science. Contextual information is often the key to structure such real world data, and we propose to use it in form of a knowledge graph. Such a knowledge graph, combined with content analysis, enhances the notion of proximity between artworks so it improves the performances in classification tasks. In this paper, we propose a novel use of a knowledge graph, that is constructed on annotated data and pseudo-labeled data. With label propagation, we boost artwork classification by training a model using a graph convolutional network, relying on the relationships between entities of the knowledge graph. Following a transductive learning framework, our experiments show that relying on a knowledge graph modeling the relations between labeled data and unlabeled data allows to achieve state-of-the-art results on multiple classification tasks on a dataset of paintings, and on a dataset of Buddha statues. Additionally, we show state-of-the-art results for the difficult case of dealing with unbalanced data, with the limitation of disregarding classes with extremely low degrees in the knowledge graph.

Vinay Damodaran, Sharanya Chakravarthy, Akshay Kumar et al.

Visual Question Answering (VQA) is of tremendous interest to the research community with important applications such as aiding visually impaired users and image-based search. In this work, we explore the use of scene graphs for solving the VQA task. We conduct experiments on the GQA dataset which presents a challenging set of questions requiring counting, compositionality and advanced reasoning capability, and provides scene graphs for a large number of images. We adopt image + question architectures for use with scene graphs, evaluate various scene graph generation techniques for unseen images, propose a training curriculum to leverage human-annotated and auto-generated scene graphs, and build late fusion architectures to learn from multiple image representations. We present a multi-faceted study into the use of scene graphs for VQA, making this work the first of its kind.

Nikolai Huckle, Noa Garcia, Yuta Nakashima

Computational art analysis has, through its reliance on classification tasks, prioritised historical datasets in which the artworks are already well sorted with the necessary annotations. Art produced today, on the other hand, is numerous and easily accessible, through the internet and social networks that are used by professional and amateur artists alike to display their work. Although this art, yet unsorted in terms of style and genre, is less suited for supervised analysis, the data sources come with novel information that may help frame the visual content in equally novel ways. As a first step in this direction, we present contempArt, a multi-modal dataset of exclusively contemporary artworks. contempArt is a collection of paintings and drawings, a detailed graph network based on social connections on Instagram and additional socio-demographic information; all attached to 442 artists at the beginning of their career. We evaluate three methods suited for generating unsupervised style embeddings of images and correlate them with the remaining data. We find no connections between visual style on the one hand and social proximity, gender, and nationality on the other.

Noa Garcia, Chentao Ye, Zihua Liu et al.

Answering questions related to art pieces (paintings) is a difficult task, as it implies the understanding of not only the visual information that is shown in the picture, but also the contextual knowledge that is acquired through the study of the history of art. In this work, we introduce our first attempt towards building a new dataset, coined AQUA (Art QUestion Answering). The question-answer (QA) pairs are automatically generated using state-of-the-art question generation methods based on paintings and comments provided in an existing art understanding dataset. The QA pairs are cleansed by crowdsourcing workers with respect to their grammatical correctness, answerability, and answers' correctness. Our dataset inherently consists of visual (painting-based) and knowledge (comment-based) questions. We also present a two-branch model as baseline, where the visual and knowledge questions are handled independently. We extensively compare our baseline model against the state-of-the-art models for question answering, and we provide a comprehensive study about the challenges and potential future directions for visual question answering on art.

Noa Garcia, Yuta Nakashima

To understand movies, humans constantly reason over the dialogues and actions shown in specific scenes and relate them to the overall storyline already seen. Inspired by this behaviour, we design ROLL, a model for knowledge-based video story question answering that leverages three crucial aspects of movie understanding: dialog comprehension, scene reasoning, and storyline recalling. In ROLL, each of these tasks is in charge of extracting rich and diverse information by 1) processing scene dialogues, 2) generating unsupervised video scene descriptions, and 3) obtaining external knowledge in a weakly supervised fashion. To answer a given question correctly, the information generated by each inspired-cognitive task is encoded via Transformers and fused through a modality weighting mechanism, which balances the information from the different sources. Exhaustive evaluation demonstrates the effectiveness of our approach, which yields a new state-of-the-art on two challenging video question answering datasets: KnowIT VQA and TVQA+.

Noa Garcia, Mayu Otani, Chenhui Chu et al.

We propose a novel video understanding task by fusing knowledge-based and video question answering. First, we introduce KnowIT VQA, a video dataset with 24,282 human-generated question-answer pairs about a popular sitcom. The dataset combines visual, textual and temporal coherence reasoning together with knowledge-based questions, which need of the experience obtained from the viewing of the series to be answered. Second, we propose a video understanding model by combining the visual and textual video content with specific knowledge about the show. Our main findings are: (i) the incorporation of knowledge produces outstanding improvements for VQA in video, and (ii) the performance on KnowIT VQA still lags well behind human accuracy, indicating its usefulness for studying current video modelling limitations.

Noa Garcia, Mayu Otani, Chenhui Chu et al.

We propose a novel video understanding task by fusing knowledge-based and video question answering. First, we introduce KnowIT VQA, a video dataset with 24,282 human-generated question-answer pairs about a popular sitcom. The dataset combines visual, textual and temporal coherence reasoning together with knowledge-based questions, which need of the experience obtained from the viewing of the series to be answered. Second, we propose a video understanding model by combining the visual and textual video content with specific knowledge about the show. Our main findings are: (i) the incorporation of knowledge produces outstanding improvements for VQA in video, and (ii) the performance on KnowIT VQA still lags well behind human accuracy, indicating its usefulness for studying current video modelling limitations.

Benjamin Renoust, Matheus Oliveira Franca, Jacob Chan et al.

While Buddhism has spread along the Silk Roads, many pieces of art have been displaced. Only a few experts may identify these works, subjectively to their experience. The construction of Buddha statues was taught through the definition of canon rules, but the applications of those rules greatly varies across time and space. Automatic art analysis aims at supporting these challenges. We propose to automatically recover the proportions induced by the construction guidelines, in order to use them and compare between different deep learning features for several classification tasks, in a medium size but rich dataset of Buddha statues, collected with experts of Buddhism art history.

Noa Garcia, Benjamin Renoust, Yuta Nakashima

In computer vision, visual arts are often studied from a purely aesthetics perspective, mostly by analysing the visual appearance of an artistic reproduction to infer its style, its author, or its representative features. In this work, however, we explore art from both a visual and a language perspective. Our aim is to bridge the gap between the visual appearance of an artwork and its underlying meaning, by jointly analysing its aesthetics and its semantics. We introduce the use of multi-modal techniques in the field of automatic art analysis by 1) collecting a multi-modal dataset with fine-art paintings and comments, and 2) exploring robust visual and textual representations in artistic images.

Noa Garcia, Benjamin Renoust, Yuta Nakashima

Automatic art analysis aims to classify and retrieve artistic representations from a collection of images by using computer vision and machine learning techniques. In this work, we propose to enhance visual representations from neural networks with contextual artistic information. Whereas visual representations are able to capture information about the content and the style of an artwork, our proposed context-aware embeddings additionally encode relationships between different artistic attributes, such as author, school, or historical period. We design two different approaches for using context in automatic art analysis. In the first one, contextual data is obtained through a multi-task learning model, in which several attributes are trained together to find visual relationships between elements. In the second approach, context is obtained through an art-specific knowledge graph, which encodes relationships between artistic attributes. An exhaustive evaluation of both of our models in several art analysis problems, such as author identification, type classification, or cross-modal retrieval, show that performance is improved by up to 7.3% in art classification and 37.24% in retrieval when context-aware embeddings are used.

Noa Garcia, George Vogiatzis

Automatic art analysis has been mostly focused on classifying artworks into different artistic styles. However, understanding an artistic representation involves more complex processes, such as identifying the elements in the scene or recognizing author influences. We present SemArt, a multi-modal dataset for semantic art understanding. SemArt is a collection of fine-art painting images in which each image is associated to a number of attributes and a textual artistic comment, such as those that appear in art catalogues or museum collections. To evaluate semantic art understanding, we envisage the Text2Art challenge, a multi-modal retrieval task where relevant paintings are retrieved according to an artistic text, and vice versa. We also propose several models for encoding visual and textual artistic representations into a common semantic space. Our best approach is able to find the correct image within the top 10 ranked images in the 45.5% of the test samples. Moreover, our models show remarkable levels of art understanding when compared against human evaluation.

Noa Garcia, George Vogiatzis

This work proposes a system for retrieving clothing and fashion products from video content. Although films and television are the perfect showcase for fashion brands to promote their products, spectators are not always aware of where to buy the latest trends they see on screen. Here, a framework for breaking the gap between fashion products shown on videos and users is presented. By relating clothing items and video frames in an indexed database and performing frame retrieval with temporal aggregation and fast indexing techniques, we can find fashion products from videos in a simple and non-intrusive way. Experiments in a large-scale dataset conducted here show that, by using the proposed framework, memory requirements can be reduced by 42.5X with respect to linear search, whereas accuracy is maintained at around 90%.

Noa Garcia, George Vogiatzis

Measuring visual similarity between two or more instances within a data distribution is a fundamental task in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the system. In this work, we explore neural networks models for learning a non-metric similarity function for instance search. We argue that non-metric similarity functions based on neural networks can build a better model of human visual perception than standard metric distances. As our proposed similarity function is differentiable, we explore a real end-to-end trainable approach for image retrieval, i.e. we learn the weights from the input image pixels to the final similarity score. Experimental evaluation shows that non-metric similarity networks are able to learn visual similarities between images and improve performance on top of state-of-the-art image representations, boosting results in standard image retrieval datasets with respect standard metric distances.