{"id":"https://openalex.org/W7135186445","doi":"https://doi.org/10.48550/arxiv.2603.12247","title":"Trust Your Critic: Robust Reward Modeling and Reinforcement Learning for Faithful Image Editing and Generation","display_name":"Trust Your Critic: Robust Reward Modeling and Reinforcement Learning for Faithful Image Editing and Generation","publication_year":2026,"publication_date":"2026-03-12","ids":{"openalex":"https://openalex.org/W7135186445","doi":"https://doi.org/10.48550/arxiv.2603.12247"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.12247","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.12247","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.12247","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5129047306","display_name":"Xiangyu Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Zhao, Xiangyu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5001609288","display_name":"Peiyuan Zhang","orcid":"https://orcid.org/0000-0003-4086-3436"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Peiyuan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128979772","display_name":"Junming Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Junming","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129079063","display_name":"Tianhao Liang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liang, Tianhao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128705322","display_name":"Yuchen Duan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Duan, Yuchen","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129073041","display_name":"Shengyuan Ding","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ding, Shengyuan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128981155","display_name":"Changyao Tian","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tian, Changyao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129014568","display_name":"Yuhang Zang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zang, Yuhang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128972085","display_name":"Junchi Yan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yan, Junchi","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5129008200","display_name":"Xue Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Xue","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":10,"corresponding_author_ids":["https://openalex.org/A5129047306"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.5514000058174133,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.5514000058174133,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12859","display_name":"Cell Image Analysis Techniques","score":0.052799999713897705,"subfield":{"id":"https://openalex.org/subfields/1304","display_name":"Biophysics"},"field":{"id":"https://openalex.org/fields/13","display_name":"Biochemistry, Genetics and Molecular Biology"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T12650","display_name":"Aesthetic Perception and Analysis","score":0.0364999994635582,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.7347999811172485},{"id":"https://openalex.org/keywords/fidelity","display_name":"Fidelity","score":0.6739000082015991},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.5877000093460083},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.529699981212616},{"id":"https://openalex.org/keywords/image-editing","display_name":"Image editing","score":0.4966999888420105},{"id":"https://openalex.org/keywords/construct","display_name":"Construct (python library)","score":0.48590001463890076}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8672999739646912},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7347999811172485},{"id":"https://openalex.org/C2776459999","wikidata":"https://www.wikidata.org/wiki/Q2119376","display_name":"Fidelity","level":2,"score":0.6739000082015991},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.593500018119812},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.5877000093460083},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.529699981212616},{"id":"https://openalex.org/C2776674983","wikidata":"https://www.wikidata.org/wiki/Q545981","display_name":"Image editing","level":3,"score":0.4966999888420105},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4909000098705292},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.48590001463890076},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.4690000116825104},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.4578000009059906},{"id":"https://openalex.org/C113364801","wikidata":"https://www.wikidata.org/wiki/Q26674","display_name":"High fidelity","level":2,"score":0.36719998717308044},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.3391000032424927},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.2676999866962433},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.2637999951839447}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.12247","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.12247","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.12247","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.12247","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Reinforcement":[0],"learning":[1],"(RL)":[2],"has":[3],"emerged":[4],"as":[5,23],"a":[6,50,126,163,205],"promising":[7],"paradigm":[8],"for":[9,64,131,175,181,208],"enhancing":[10],"image":[11,66],"editing":[12,85,132,176],"and":[13,31,61,68,89,106,109,115,133,177,191,210,222],"text-to-image":[14],"(T2I)":[15],"generation.":[16,182],"However,":[17],"current":[18],"reward":[19,56,112,166],"models,":[20,221],"which":[21],"act":[22],"critics":[24,156],"during":[25],"RL,":[26],"often":[27],"suffer":[28],"from":[29],"hallucinations":[30],"assign":[32],"noisy":[33],"scores,":[34],"inherently":[35],"misguiding":[36],"the":[37,104,158],"optimization":[38],"process.":[39],"In":[40],"this":[41,185],"paper,":[42],"we":[43,71,83,102,123,161],"present":[44],"FIRM":[45,201],"(Faithful":[46],"Image":[47],"Reward":[48],"Modeling),":[49],"comprehensive":[51,127],"framework":[52],"that":[53,117,138,168,200],"develops":[54],"robust":[55],"models":[57,113,140,189],"to":[58,77,148,152],"provide":[59],"accurate":[60],"reliable":[62],"guidance":[63],"faithful":[65],"generation":[67,92,134],"editing.":[69],"First,":[70],"design":[72],"tailored":[73],"data":[74],"curation":[75],"pipelines":[76],"construct":[78],"high-quality":[79],"scoring":[80],"datasets.":[81],"Specifically,":[82],"evaluate":[84],"using":[86],"both":[87],"execution":[88],"consistency,":[90],"while":[91],"is":[93],"primarily":[94],"assessed":[95],"via":[96],"instruction":[97,211],"following.":[98],"Using":[99],"these":[100,120,155],"pipelines,":[101],"collect":[103],"FIRM-Edit-370K":[105],"FIRM-Gen-293K":[107],"datasets,":[108,220],"train":[110],"specialized":[111],"(FIRM-Edit-8B":[114],"FIRM-Gen-8B)":[116],"accurately":[118],"reflect":[119],"criteria.":[121],"Second,":[122],"introduce":[124],"FIRM-Bench,":[125],"benchmark":[128],"specifically":[129],"designed":[130],"critics.":[135],"Evaluations":[136],"demonstrate":[137,199],"our":[139,187,219],"achieve":[141,193],"superior":[142],"alignment":[143],"with":[144],"human":[145],"judgment":[146],"compared":[147],"existing":[149,214],"metrics.":[150],"Furthermore,":[151],"seamlessly":[153],"integrate":[154],"into":[157],"RL":[159],"pipeline,":[160],"formulate":[162],"novel":[164],"\"Base-and-Bonus\"":[165],"strategy":[167],"balances":[169],"competing":[170],"objectives:":[171],"Consistency-Modulated":[172],"Execution":[173],"(CME)":[174],"Quality-Modulated":[178],"Alignment":[179],"(QMA)":[180],"Empowered":[183],"by":[184],"framework,":[186],"resulting":[188],"FIRM-Qwen-Edit":[190],"FIRM-SD3.5":[192],"substantial":[194],"performance":[195],"breakthroughs.":[196],"Comprehensive":[197],"experiments":[198],"mitigates":[202],"hallucinations,":[203],"establishing":[204],"new":[206],"standard":[207],"fidelity":[209],"adherence":[212],"over":[213],"general":[215],"models.":[216],"All":[217],"of":[218],"code":[223],"have":[224],"been":[225],"publicly":[226],"available":[227],"at":[228],"https://firm-reward.github.io.":[229]},"counts_by_year":[],"updated_date":"2026-03-14T06:46:50.379900","created_date":"2026-03-14T00:00:00"}