Answering complex visual questions like `Which red furniture can be used for sitting?' requires multi-step reasoning, including object recognition, attribute filtering, and relational understanding. Recent work improves interpretability in multimodal large language models (MLLMs) by decomposing tasks into sub-task programs, but these methods are computationally expensive and less accurate due to poor adaptation to target data. To address this, we introduce VISTAR (Visually Interpretable Subtask-Aware Reasoning Model), a subtask-driven training framework that enhances both interpretability and reasoning by generating textual and visual explanations within MLLMs. Instead of relying on external models, VISTAR fine-tunes MLLMs to produce structured Subtask-of-Thought rationales (step-by-step reasoning sequences). Experiments on two benchmarks show that VISTAR consistently improves reasoning accuracy while maintaining interpretability. Our code and dataset will be available atthis https URL.
View on arXiv@article{cheng2025_2505.08084,
title={ Visually Interpretable Subtask Reasoning for Visual Question Answering },
author={ Yu Cheng and Arushi Goel and Hakan Bilen },
journal={arXiv preprint arXiv:2505.08084},
year={ 2025 }
}