Analyzing the Behavior of Visual Question Answering Models
Open accessAbstract
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).
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Certain datasets and models are biased to the most dominant modality, sometimes ignoring others completely.
“and that certain datasets and models are biased to the most dominant modality [22, 59], sometimes ignoring others completely [3].”
From MultiBench: Multiscale Benchmarks for Multimodal Representation Learning (2021) · cites this paper as [3]
Observations that models rely on unimodal statistics or that datasets and models are biased to the dominant modality are currently only conducted on specific datasets and models without testing their generalization to others.
“These observations are currently only conducted on specific datasets and models without testing their generalization to others, a shortcoming we hope to solve using MULTIBENCH which enables scalable analysis over modalities, tasks, and models.”
From MultiBench: Multiscale Benchmarks for Multimodal Representation Learning (2021) · cites this paper as [3]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Efficient Estimation of Word Representations in Vector Space
Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013) · arXiv · doi:10.48550/arxiv.1301.3781 · Cited by 26,568 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.74 · Cited by 24,513 · see citing works
Self-citationGrad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2020) · International Journal of Computer Vision · doi:10.1007/s11263-019-01228-7 · Cited by 6,909 · see citing works
Self-citationMultimodal Machine Learning: A Survey and Taxonomy
Baltrusaitis, T., Ahuja, C., & Morency, L.-P. (2019) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2018.2798607 · Cited by 5,018 · see citing works
Shortcut learning in deep neural networks
Geirhos, R., Jacobsen, J.-H., Michaelis, C., Zemel, R., Brendel, W., Bethge, M., & Wichmann, F. A. (2020) · Nature Machine Intelligence · doi:10.1038/s42256-020-00257-z · Cited by 2,224 · see citing works
Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference
McCoy, R. T., Pavlick, E., & Linzen, T. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1334 · Cited by 944 · see citing works
Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering
Goyal, Y., Khot, T., Agrawal, A., Summers-Stay, D., Batra, D., & Parikh, D. (2019) · International Journal of Computer Vision · doi:10.1007/s11263-018-1116-0 · Cited by 487 · see citing works
Self-citationThe hateful memes challenge: Detecting hate speech in multimodal memes
Kiela, D., Firooz, H., Mohan, A., Goswami, V., Singh, A., Ringshia, P., & Testuggine, D. (2020) · arXiv · doi:10.48550/arxiv.2005.04790 · Cited by 352 · see citing works
Did the Model Understand the Question?
Mudrakarta, P. K., Taly, A., Sundararajan, M., & Dhamdhere, K. (2018) · Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p18-1176 · Cited by 166 · see citing works
Exact and Consistent Interpretation for Piecewise Linear Neural Networks: A Closed Form Solution
Chu, L., Hu, X., Hu, J., Wang, L., & Pei, J. (2018) · Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining · doi:10.1145/3219819.3220063 · Cited by 87 · see citing works
CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning
Johnson, J., Hariharan, B., van der Maaten, L., Fei-Fei, L., Zitnick, C. L., & Girshick, R. (2016) · arXiv · doi:10.48550/arxiv.1612.06890 · Cited by 40 · see citing works
VisQA: X-raying Vision and Language Reasoning in Transformers
Jaunet, T., Kervadec, C., Vuillemot, R., Antipov, G., Baccouche, M., & Wolf, C. (2022) · IEEE Transactions on Visualization and Computer Graphics · doi:10.1109/tvcg.2021.3114683 · Cited by 30 · see citing works
MultiBench: Multiscale Benchmarks for Multimodal Representation Learning
Liang, P. P., Lyu, Y., Fan, X., Wu, Z., Cheng, Y., Wu, J., Chen, L., Wu, P., Lee, M. A., Zhu, Y., et al. (2021) · arXiv · doi:10.48550/arxiv.2107.07502 · Cited by 85 · see citing works
“They have shown that certain models do not actually learn cross-modal interactions but rather rely on ensembles of unimodal statistics [68] and that certain datasets and models are biased to the most dominant modality [22, 59], sometimes ignoring others completely [3].”
BackgroundCLEVR-X: A Visual Reasoning Dataset for Natural Language Explanations
Salewski, L., Koepke, A. S., Lensch, H. P. A., & Akata, Z. (2022) · Lecture Notes in Computer Science · doi:10.1007/978-3-031-04083-2_5 · Cited by 20 · see citing works
- Authors
- Agrawal, A., Batra, D., & Parikh, D.
- Affiliations
- Virginia Tech (US)
- Venue
- Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing
- Published
- 2016
- Pages
- 1955-1960
- OpenAlex
- W2463267937
- Cited by
- 297
- References
- 1
- Citations by year
- 2026: 12 · 2025: 26 · 2024: 21 · 2023: 31 · 2022: 32 · 2021: 45 · 2020: 42 · 2019: 34 · 2018: 38 · 2017: 15 · 2016: 1
- FWCI
- 9.96
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Conference on Empirical Methods in Natural Language Processing (EMNLP) (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Science Foundation (CAREER)
- Nvidia
- Office of Naval Research (N00014-14-1-0679, N00014)
- Army Research Office
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Visual Attention and Saliency Detection
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Keywords
- visual question answering, model behavior analysis, attention-based models, myopia, early prediction
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
