VQA4CIR: Boosting Composed Image Retrieval with Visual Question Answering
This work addresses retrieval inconsistencies in CIR, which is important for applications like e-commerce and content search, but it is incremental as it builds on existing CIR methods with a plug-in approach.
The paper tackles the problem of failure retrieval results in Composed Image Retrieval (CIR) that are inconsistent with captions by introducing VQA4CIR, a post-processing method using Visual Question Answering to identify and re-rank such images, resulting in improved performance over state-of-the-art methods on CIRR and Fashion-IQ datasets.
Albeit progress has been made in Composed Image Retrieval (CIR), we empirically find that a certain percentage of failure retrieval results are not consistent with their relative captions. To address this issue, this work provides a Visual Question Answering (VQA) perspective to boost the performance of CIR. The resulting VQA4CIR is a post-processing approach and can be directly plugged into existing CIR methods. Given the top-C retrieved images by a CIR method, VQA4CIR aims to decrease the adverse effect of the failure retrieval results being inconsistent with the relative caption. To find the retrieved images inconsistent with the relative caption, we resort to the "QA generation to VQA" self-verification pipeline. For QA generation, we suggest fine-tuning LLM (e.g., LLaMA) to generate several pairs of questions and answers from each relative caption. We then fine-tune LVLM (e.g., LLaVA) to obtain the VQA model. By feeding the retrieved image and question to the VQA model, one can find the images inconsistent with relative caption when the answer by VQA is inconsistent with the answer in the QA pair. Consequently, the CIR performance can be boosted by modifying the ranks of inconsistently retrieved images. Experimental results show that our proposed method outperforms state-of-the-art CIR methods on the CIRR and Fashion-IQ datasets.