Exploring Question Decomposition for Zero-Shot VQA

Zaid Khan, Vijay Kumar b g, Samuel Schulter, Manmohan Chandraker, Yun Fu

Keywords: visual question answering, in-context learning, vision-language

TL;DR: Can recent instruction-tuned models improve their own predictions on 0-shot VQA tasks by breaking down the question? Yes.

Abstract:

Visual question answering (VQA) has traditionally been treated as a single-step task where each question receives the same amount of effort, unlike natural human question-answering strategies. We explore a question decomposition strategy for VQA to overcome this limitation. We probe the ability of recently developed large vision-language models to use human-written decompositions and produce their own decompositions of visual questions, finding they are capable of learning both tasks from demonstrations alone.
However, we show that naive application of model-written decompositions can hurt performance.
We introduce a model-driven selective decomposition approach for second-guessing predictions and correcting errors, and validate its effectiveness on eight VQA tasks across three domains, showing consistent improvements in accuracy, including improvements of >20% on medical VQA datasets and boosting the zero-shot performance of BLIP-2 above chance on a VQA reformulation of the challenging Winoground task. Project Site: https://zaidkhan.me/decomposition-0shot-vqa/

Submission Number: 2161