{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/arxiv-2601-22920","title":"Q-Hawkeye: Reliable Visual Policy Optimization for Image Quality Assessment","arxiv_id":"2601.22920","date":"2026-01-30","proceeding":null,"authors":["Wulin Xie","Rui Dai","Ruidong Ding","Kaikui Liu","Xiangxiang Chu","Xinwen Hou","Jie Wen"],"abstract":"Image Quality Assessment (IQA) predicts perceptual quality scores consistent with human judgments. Recent RL-based IQA methods built on MLLMs focus on generating visual quality descriptions and scores, ignoring two key reliability limitations: (i) although the model's prediction stability varies significantly across training samples, existing GRPO-based methods apply uniform advantage weighting, thereby amplifying noisy signals from unstable samples in gradient updates; (ii) most works emphasize text-grounded reasoning over images while overlooking the model's visual perception ability of image content. In this paper, we propose Q-Hawkeye, an RL-based reliable visual policy optimization framework that redesigns the learning signal through unified Uncertainty-Aware Dynamic Optimization and Perception-Aware Optimization. Q-Hawkeye estimates predictive uncertainty using the variance of predicted scores across multiple rollouts and leverages this uncertainty to reweight each sample's update strength, stabilizing policy optimization. To strengthen perceptual reliability, we construct paired inputs of degraded images and their original images and introduce an Implicit Perception Loss that constrains the model to ground its quality judgments in genuine visual evidence. Extensive experiments demonstrate that Q-Hawkeye outperforms state-of-the-art methods and generalizes better across multiple datasets. Our dataset and code are available at https://github.com/AMAP-ML/Q-Hawkeye.","url_abs":"https://arxiv.org/abs/2601.22920","url_pdf":"https://arxiv.org/pdf/2601.22920","source":{"archive":null,"snapshot":"2025-07-28","note":"not in the Papers with Code archive (frozen at the snapshot)","row_kind":"graph","title_abstract_authors_date":"arXiv metadata, CC0 1.0 (https://info.arxiv.org/help/license)"},"code_links":[],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=2601.22920","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2601.22920"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"mentioned_in_github":null,"is_official":null,"provenance":"deterministic:regex_extraction","mentioned_in_paper":null,"url":"https://github.com/AMAP-ML/Q-Hawkeye","reach":null}],"summary":{"ran_fixture":1,"unverified":13},"by_repo_kind":{"found_in_text":{"samples":14,"ran":1,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":14,"samples":[{"code_sha256_prefix":"3f31b48fd862935d","entry":"get_image_data_url","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/virft/local_scripts/create_vision_cot_data.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/virft/local_scripts/create_vision_cot_data.py","link_basis":"plan_row","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"RAISES","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"3f31b48fd862935d"}},{"code_sha256_prefix":"39948c4c2e2232d9","entry":"calculate_iou","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/virft/src/open_r1/grpo.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/virft/src/open_r1/grpo.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"39948c4c2e2232d9"}},{"code_sha256_prefix":"7ed1cffd12a5fc04","entry":"clean_text","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/virft/src/open_r1/grpo_am.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/virft/src/open_r1/grpo_am.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"7ed1cffd12a5fc04"}},{"code_sha256_prefix":"0e7aa9a2af82e857","entry":"create_comparison_prompt","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/Dataset/Degradation_Dataset/VLM_filter.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/Dataset/Degradation_Dataset/VLM_filter.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"0e7aa9a2af82e857"}},{"code_sha256_prefix":"587a00453d7c872e","entry":"encode_image","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/Dataset/Degradation_Dataset/VLM_filter.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/Dataset/Degradation_Dataset/VLM_filter.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"587a00453d7c872e"}},{"code_sha256_prefix":"93ad7eab916a291e","entry":"extract_bbox","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/virft/src/open_r1/grpo.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/virft/src/open_r1/grpo.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"93ad7eab916a291e"}},{"code_sha256_prefix":"41c548a25e90eb39","entry":"extract_problem_solution","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/virft/local_scripts/prepare_hf_data.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/virft/local_scripts/prepare_hf_data.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"41c548a25e90eb39"}},{"code_sha256_prefix":"72a415a34fdb8943","entry":"generate_html","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/Dataset/Degradation_Dataset/Human_filter.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/Dataset/Degradation_Dataset/Human_filter.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"72a415a34fdb8943"}},{"code_sha256_prefix":"8c7273a7482d604e","entry":"image_to_base64","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/Dataset/Degradation_Dataset/Human_filter.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/Dataset/Degradation_Dataset/Human_filter.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"8c7273a7482d604e"}},{"code_sha256_prefix":"bdcc94560fb87f30","entry":"load_image_from_path","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/virft/local_scripts/prepare_hf_data.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/virft/local_scripts/prepare_hf_data.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"bdcc94560fb87f30"}},{"code_sha256_prefix":"47c1b2bda826b75a","entry":"process_json_data","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/Dataset/Degradation_Dataset/Human_filter.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/Dataset/Degradation_Dataset/Human_filter.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"47c1b2bda826b75a"}},{"code_sha256_prefix":"478ce302c9c491d6","entry":"repetition_reward","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/virft/src/open_r1/grpo_am.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/virft/src/open_r1/grpo_am.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"478ce302c9c491d6"}},{"code_sha256_prefix":"5420f5729319b51d","entry":"score_reward","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/virft/src/open_r1/grpo_am.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/virft/src/open_r1/grpo_am.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"5420f5729319b51d"}},{"code_sha256_prefix":"99496fea41fb9d81","entry":"sort_and_calculate_iou","repo":"AMAP-ML/Q-Hawkeye","repo_kind":"found_in_text","path":"src/virft/src/open_r1/grpo.py","file_url":"https://github.com/AMAP-ML/Q-Hawkeye/blob/HEAD/src/virft/src/open_r1/grpo.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"99496fea41fb9d81"}}]},"arxiv_metadata":{"licence":"arXiv metadata, CC0 1.0 (https://info.arxiv.org/help/license)","fields":["title","abstract","authors","date"],"primary_category":"cs.CV","source":"arxiv_2026.jsonl"},"syntology_extracted_results":null}