RUBi: Reducing Unimodal Biases in Visual Question Answering

Abstract

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

arXiv commentNeurIPS 2019 http://papers.nips.cc/paper/8371-rubi-reducing-unimodal-biases-for-visual-question-answering

References in corpus (6)

• Efficient Estimation of Word Representations in Vector Space
• Hierarchical Question-Image Co-Attention for Visual Question Answering
• Beyond Bilinear: Generalized Multimodal Factorized High-order Pooling for Visual Question Answering
• Women also Snowboard: Overcoming Bias in Captioning Models
• Robot Learning in Homes: Improving Generalization and Reducing Dataset Bias
• Blindfold Baselines for Embodied QA

Cited by in corpus (20)

• Multimodal Intelligence: Representation Learning, Information Fusion, and Applications
• Multimodal Co-learning: Challenges, Applications with Datasets, Recent Advances and Future Directions
• On the Value of Out-of-Distribution Testing: An Example of Goodhart's Law
• Beyond Question-Based Biases: Assessing Multimodal Shortcut Learning in Visual Question Answering
• Learning from others' mistakes: Avoiding dataset biases without modeling them
• LPF: A Language-Prior Feedback Objective Function for De-biased Visual Question Answering
• Exploring the State of the Art in Legal QA Systems
• Unshuffling Data for Improved Generalization
• Unbiased Scene Graph Generation from Biased Training
• Gender and Racial Bias in Visual Question Answering Datasets
• Counterfactual Samples Synthesizing for Robust Visual Question Answering
• Counterfactual VQA: A Cause-Effect Look at Language Bias
• Redundancy-aware Transformer for Video Question Answering
• Using Positive Matching Contrastive Loss with Facial Action Units to mitigate bias in Facial Expression Recognition
• Single-Modal Entropy based Active Learning for Visual Question Answering
• Eliminating the Language Bias for Visual Question Answering with fine-grained Causal Intervention
• Generating Rationales in Visual Question Answering
• Overcoming Language Priors with Self-supervised Learning for Visual Question Answering
• Disentangling private classes through regularization
• Calibrating Concepts and Operations: Towards Symbolic Reasoning on Real Images