{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,31]],"date-time":"2026-03-31T13:54:38Z","timestamp":1774965278303,"version":"3.50.1"},"reference-count":0,"publisher":"Association for the Advancement of Artificial Intelligence (AAAI)","issue":"27","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["AAAI"],"abstract":"<jats:p>Reinforcement fine-tuning (RFT) is a proliferating paradigm for LMM training. \nAnalogous to high-level reasoning tasks, RFT is similarly applicable to low-level vision domains, including image quality assessment (IQA). Existing RFT-based IQA methods typically use rule-based output rewards to verify the model's rollouts but provide no reward supervision for the \"think\u201d process, leaving its correctness and efficacy uncontrolled. Furthermore, these methods typically fine-tune directly on downstream IQA tasks without explicitly enhancing the model\u2019s native low-level visual quality perception, which may constrain its performance upper bound. In response to these gaps, we propose the multi\u2010stage RFT IQA framework (Refine-IQA). In Stage-1, we build the Refine-Perception-20K dataset (with 12 main distortions, 20,907 locally-distorted images, and over 55K RFT samples) and design multi-task reward functions to strengthen the model\u2019s visual quality perception. In Stage-2, targeting the quality scoring task, we introduce a probability difference reward involved strategy for \"think\" process supervision. The resulting Refine-IQA Series Models achieve outstanding performance on both perception and scoring tasks\u2014and, notably, our paradigm activates a robust \"think\u201d (quality interpretating) capability that also attains exceptional results on the corresponding quality interpreting benchmark.<\/jats:p>","DOI":"10.1609\/aaai.v40i27.39387","type":"journal-article","created":{"date-parts":[[2026,3,18]],"date-time":"2026-03-18T01:31:33Z","timestamp":1773797493000},"page":"22301-22309","source":"Crossref","is-referenced-by-count":3,"title":["Refine-IQA: Multi-Stage Reinforcement Finetuning for Perceptual Image Quality Assessment"],"prefix":"10.1609","volume":"40","author":[{"given":"Ziheng","family":"Jia","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Jiaying","family":"Qian","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Zicheng","family":"Zhang","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Zijian","family":"Chen","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Xiongkuo","family":"Min","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"9382","published-online":{"date-parts":[[2026,3,14]]},"container-title":["Proceedings of the AAAI Conference on Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/ojs.aaai.org\/index.php\/AAAI\/article\/download\/39387\/43348","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/ojs.aaai.org\/index.php\/AAAI\/article\/download\/39387\/43348","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,18]],"date-time":"2026-03-18T01:31:33Z","timestamp":1773797493000},"score":1,"resource":{"primary":{"URL":"https:\/\/ojs.aaai.org\/index.php\/AAAI\/article\/view\/39387"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,3,14]]},"references-count":0,"journal-issue":{"issue":"27","published-online":{"date-parts":[[2026,3,17]]}},"URL":"https:\/\/doi.org\/10.1609\/aaai.v40i27.39387","relation":{},"ISSN":["2374-3468","2159-5399"],"issn-type":[{"value":"2374-3468","type":"electronic"},{"value":"2159-5399","type":"print"}],"subject":[],"published":{"date-parts":[[2026,3,14]]}}}