RUBi: Reducing Unimodal Biases in Visual Question Answering
This addresses a critical issue for VQA systems, making them more robust and suitable for real-world applications by reducing reliance on biases, though it is an incremental improvement over existing debiasing methods.
The paper tackles the problem of unimodal biases in Visual Question Answering (VQA) models, which rely on statistical regularities between questions and answers instead of using image information, leading to poor generalization. The proposed RUBi strategy reduces these biases by leveraging a question-only model to adjust the loss dynamically, achieving state-of-the-art results on the VQA-CP v2 dataset.
Visual Question Answering (VQA) is the task of answering questions about an image. Some VQA models often exploit unimodal biases to provide the correct answer without using the image information. As a result, they suffer from a huge drop in performance when evaluated on data outside their training set distribution. This critical issue makes them unsuitable for real-world settings. We propose RUBi, a new learning strategy to reduce biases in any VQA model. It reduces the importance of the most biased examples, i.e. examples that can be correctly classified without looking at the image. It implicitly forces the VQA model to use the two input modalities instead of relying on statistical regularities between the question and the answer. We leverage a question-only model that captures the language biases by identifying when these unwanted regularities are used. It prevents the base VQA model from learning them by influencing its predictions. This leads to dynamically adjusting the loss in order to compensate for biases. We validate our contributions by surpassing the current state-of-the-art results on VQA-CP v2. This dataset is specifically designed to assess the robustness of VQA models when exposed to different question biases at test time than what was seen during training. Our code is available: github.com/cdancette/rubi.bootstrap.pytorch