{"id":"W4417423761","doi":"10.1109/iccv51701.2025.00061","title":"NegRefine: Refining Negative Label-Based Zero-Shot OOD Detection","year":2025,"lang":"en","type":"article","venue":"","topic":"Multimodal Machine Learning Applications","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"National Research Council Canada; Simon Fraser University","funders":"","keywords":"Subcategory; Set (abstract data type); Lexicon; Matching (statistics); Face (sociological concept); Word (group theory); Noun; Function (biology)","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002173435,0.0001378571,0.0001291237,0.0002017655,0.0002721917,0.0001381456,0.0006162708,0.00006923442,0.00002692091],"category_scores_gemma":[0.0002364326,0.0001280722,0.00004387052,0.001080691,0.00003525233,0.0001821321,0.0001630562,0.0002698277,0.0001521291],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007553449,"about_ca_system_score_gemma":0.0001056682,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009536653,"about_ca_topic_score_gemma":0.0001234399,"domain_scores_codex":[0.9988812,0.00008646631,0.000206735,0.0004436917,0.0001588205,0.0002231531],"domain_scores_gemma":[0.9987665,0.0003438736,0.00007843674,0.0006418729,0.0001114919,0.00005785973],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00002403714,0.0002441303,0.00578795,0.00004170215,0.00005103774,0.000003229717,0.0003125359,0.008365888,0.04127689,0.3075673,0.0009501929,0.6353751],"study_design_scores_gemma":[0.0009021735,0.00008245138,0.03650285,0.00003911751,0.00001102221,0.000001682526,0.00001077042,0.9130164,0.03857751,0.006012674,0.004607942,0.0002353966],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02407132,0.00001733141,0.9286385,0.004337667,0.0001693516,0.0001118891,8.504439e-7,0.0006905659,0.04196249],"genre_scores_gemma":[0.8326239,5.846152e-7,0.1645018,0.00122846,0.00001584253,0.0000747942,0.000002080835,0.00000670118,0.001545783],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9046505,"threshold_uncertainty_score":0.5222632,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02047313720632901,"score_gpt":0.3049961903917769,"score_spread":0.2845230531854479,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}