Adversarial VQA: A New Benchmark for Evaluating the Robustness of VQA Models
Authors:
- 1 October 2021
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2022-2031
Abstract
Benefiting from large-scale pre-training, we have witnessed significant performance boost on the popular Visual Question Answering (VQA) task. Despite rapid progress, it remains unclear whether these state-of-the-art (SOTA) models are robust when encountering examples in the wild. To study this, we introduce Adversarial VQA, a new large-scale VQA benchmark, collected iteratively via an adversarial human-and-model-in-the-loop procedure. Through this new benchmark, we discover several interesting findings. (i) Surprisingly, we find that during dataset collection, non-expert annotators can easily attack SOTA VQA models successfully. (ii) Both large-scale pre-trained models and adversarial training methods achieve far worse performance on the new benchmark than over standard VQA v2 dataset, revealing the fragility of these models while demonstrating the effectiveness of our adversarial dataset. (iii) When used for data augmentation, our dataset can effectively boost model performance on other robust VQA benchmarks. We hope our Adversarial VQA dataset can shed new light on robustness study in the community and serve as a valuable benchmark for future work.Keywords
All Related Versions
This publication has 32 references indexed in Scilit:
- Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering2018
- Semantically Equivalent Adversarial Rules for Debugging NLP models2018
- A Broad-Coverage Challenge Corpus for Sentence Understanding through Inference2018
- Annotation Artifacts in Natural Language Inference Data2018
- Hypothesis Only Baselines in Natural Language Inference2018
- Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering2017
- Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image AnnotationsInternational Journal of Computer Vision, 2017
- VQA: Visual Question Answering2015
- Deep visual-semantic alignments for generating image descriptions2015
- A large annotated corpus for learning natural language inference2015