Visual Grounding Methods for VQA are Working for the Wrong Reasons!
This work addresses the issue of spurious correlations in VQA for researchers, revealing that current visual grounding methods are ineffective, which is incremental as it builds on existing bias mitigation efforts.
The paper tackles the problem that Visual Question Answering (VQA) methods often rely on dataset biases rather than true visual grounding, and finds that recent bias mitigation methods improve performance due to regularization effects, not better grounding, with a simpler proposed scheme achieving near state-of-the-art results on VQA-CPv2.
Existing Visual Question Answering (VQA) methods tend to exploit dataset biases and spurious statistical correlations, instead of producing right answers for the right reasons. To address this issue, recent bias mitigation methods for VQA propose to incorporate visual cues (e.g., human attention maps) to better ground the VQA models, showcasing impressive gains. However, we show that the performance improvements are not a result of improved visual grounding, but a regularization effect which prevents over-fitting to linguistic priors. For instance, we find that it is not actually necessary to provide proper, human-based cues; random, insensible cues also result in similar improvements. Based on this observation, we propose a simpler regularization scheme that does not require any external annotations and yet achieves near state-of-the-art performance on VQA-CPv2.