{"id":"W7131067696","doi":"10.1109/iccvw69036.2025.00597","title":"AIM 2025 Challenge on Screen-Content Video Quality Assessment: Methods and Results","year":2025,"lang":"","type":"article","venue":"","topic":"Image and Video Quality Assessment","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Alexander von Humboldt-Stiftung","keywords":"Video quality; Quality (philosophy); Set (abstract data type); Subjective video quality; Source code; Frame (networking); Videoconferencing","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":["metaepi_narrow","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.01619866,0.0009119742,0.001383704,0.0005884384,0.0008846251,0.001554828,0.001833999,0.0004546823,0.0001615492],"category_scores_gemma":[0.001117158,0.0008326416,0.0004210998,0.001227529,0.0003892379,0.001338462,0.002383249,0.001202566,0.00008155216],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004330864,"about_ca_system_score_gemma":0.001005354,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002388818,"about_ca_topic_score_gemma":0.0002901684,"domain_scores_codex":[0.9860807,0.005506114,0.003079892,0.002854286,0.001202984,0.001276005],"domain_scores_gemma":[0.9912772,0.003622759,0.0007728399,0.003220721,0.0006306955,0.0004758151],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002699993,0.001077686,0.00007603512,0.0002420051,0.0002708083,0.00002238907,0.0007924708,0.00001494565,0.0003091298,0.5090138,0.00492646,0.4829842],"study_design_scores_gemma":[0.02733009,0.008405379,0.182965,0.004575723,0.0007661927,0.00002022519,0.008119789,0.3739437,0.02413865,0.09691592,0.2667866,0.006032758],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0004683485,0.001870675,0.7381678,0.05410429,0.001664963,0.0009513778,0.00005645656,0.0001906869,0.2025254],"genre_scores_gemma":[0.3876918,0.003602511,0.5291665,0.01923515,0.0003513147,0.0001405792,0.00003337389,0.00004595655,0.05973285],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4769515,"threshold_uncertainty_score":0.9994817,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2289077325429648,"score_gpt":0.4907456582736023,"score_spread":0.2618379257306375,"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."}}