Analyzing the Behavior of Visual Question Answering Models

By Aishwarya Agrawal, Dhruv Batra, Devi Parikh. 2016

344 citations recorded by the source corpus.

Recently, a number of deep-learning based models have been proposed for the task of Visual Question Answering (VQA). The performance of most models is clustered around 60-70%. In this paper we propose systematic methods to analyze the behavior of these models as a first step towards recognizing their strengths and weaknesses, and identifying the most fruitful directions for progress. We analyze two models, one each from two major classes of VQA models -- with-attention and without-attention and show the similarities and differences in the behavior of these models. We also analyze the winning entry of the VQA Challenge 2016. Our behavior analysis reveals that despite recent progress, today's VQA models are "myopic" (tend to fail on sufficiently novel instances), often "jump to conclusions" (converge on a predicted answer after 'listening' to just half the question), and are "stubborn" (do not change their answers across images).

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Natural Language Processing

Citations in the corpus

Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference (2019, background): “Similar heuristics arise in visual question answering systems (Agrawal et al., 2016).”
Annotation Artifacts in Natural Language Inference Data (2018, background): “Agrawal et al. (2016); Jabri et al.”
Shortcut learning in deep neural networks (2020, background): “Feature combination crucially depends on shortcut features like word length [59, 60, 14, 61], and consequently leads to a severe lack of robustness such as an inability to generalise to more challenging test conditions [62, 63, 64, 65].”
The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memes (2020, neutral): “Similar problems were found in VQA [3], where simple baselines without sophisticated multimodal understanding performed remarkably well [94, 35, 1, 26] and in multimodal machine translation [18, 74], where the image was found to matter relatively little [13, 17, 7].”
ShapeWorld - A new test methodology for multimodal language understanding (2017, methodology): “We evaluate several multimodal deep neural network architectures that were recently proposed for VQA (Goyal et al., 2016; Agrawal et al., 2016; Jabri et al., 2016; Antol et al., 2015; Ren et al., 2015).”
Visual Dialog (2016, background): “As analyzed in [3,19,69], this leads to a particular bias in the questions – people only ask ‘Is there a clock-tower in the picture?’ on pictures actually containing clock towers.”
Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (2016, background): “This phenomenon has been observed in image captioning [6] as well as visual question answering [47, 49, 16, 18, 1].”
GQA: A New Dataset for Real-World Visual Reasoning and Compositional Question Answering (2019, background): “VQA models, which become heavily reliant upon such statistical biases and tendencies within the answer distribution to largely circumvent the need for true visual scene understanding [1, 9, 13, 7].”
Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization (2016, background): “visual question answering [3]), the goal of transparency and explanations is to identify the failure modes [1, 17], thereby helping researchers focus their efforts on the most fruitful research directions.”
What Makes Reading Comprehension Questions Easier? (2018, background): “In visual question answering, Agrawal et al. (2016) analyzed the behavior of models with the variable length of the first question words.”
Stacking with Auxiliary Features for Visual Question Answering (2018, background): “Visual Question Answering (VQA), the task of addressing open-ended questions about images (Ma-linowski and Fritz, 2014; Antol et al., 2015), has attracted signiﬁcant attention in recent years (Andreas et al., 2016a; Goyal et al., 2016; Agrawal et al., 2016; Teney et al., 2017).”
GuessWhat?! Visual Object Discovery through Multi-modal Dialogue (2016, background): “An extensive body of work has followed from this publication, largely building on the image captioning literature [3, 26, 37, 44].”
Errudite: Scalable, Reproducible, and Testable Error Analysis (2019, background): “Counterfactual attacks to models have taken various forms, e.g., by adding distracting sentences to the context in MC (Jia and Liang, 2017), or feeding partial questions or wrong images into models (Agrawal et al., 2016; Mudrakarta et al., 2018; Feng et al., 2018).”
GQA: a new dataset for compositional question answering over real-world images (2019, background): “They memorize the precise answer distributions of different questions to handle them with relative ease, while only glancing over the provided images and at times not even considering them, let alone understanding their content [1, 7].”
Taking a HINT: Leveraging Explanations to Make Vision and Language Models More Grounded (2019, background): “Many popular and well-performing models for multimodal, vision-and-language tasks exhibit poor visual grounding – failing to appropriately associate words or phrases with the image regions they denote and relying instead on superficial linguistic correlations [2, 1, 39, 11, 13].”