Be flexible! learn to debias by sampling and prompting for robust visual question answering

Open Webpage

Jin Liu, Chongfeng Fan, Fengyu Zhou, Huijuan Xu

Published: 01 Jan 2023, Last Modified: 14 Jul 2025Inf. Process. Manag. 2023EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•A debiasing model to overcome both the visual shortcut bias and the language distribution bias in VQA.•Incorporating the prompt learning to augment the question context.•The method is model-agnostic and can improve various VQA backbone performance.