{"id":"W3206421219","doi":"10.18280/ts.400214","title":"Statistical Evaluation of Video Summarization Models from an Empirical Perspective","year":2023,"lang":"en","type":"article","venue":"Traitement du signal","topic":"Video Analysis and Summarization","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Automatic summarization; Perspective (graphical); Computer science; Empirical research; Artificial intelligence; Statistics; Mathematics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001278808,0.000118975,0.0001917885,0.0002126533,0.00009673461,0.0001017714,0.0003258797,0.00005799204,0.0002816711],"category_scores_gemma":[0.00008338034,0.0001119577,0.00006113158,0.000767609,0.00003660737,0.0008017832,0.00007626558,0.00006462116,0.00002633192],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001282149,"about_ca_system_score_gemma":0.0001541166,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002334316,"about_ca_topic_score_gemma":0.00008193438,"domain_scores_codex":[0.9974241,0.0004101874,0.0003853078,0.0004365831,0.001163341,0.0001805113],"domain_scores_gemma":[0.9987459,0.0001489303,0.0001226657,0.0002796153,0.0006122327,0.00009063882],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00006271412,0.0006186097,0.01226234,0.00001249627,0.0002499228,0.000008209872,0.01159648,0.4710609,0.005600971,0.4229413,0.003151179,0.07243485],"study_design_scores_gemma":[0.0004532905,0.00009586255,0.03588706,0.000007211508,0.00007847233,1.71871e-7,0.0002505586,0.8636484,0.000372389,0.09907807,0.00002269153,0.0001058162],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09960855,0.00001996479,0.8987605,0.0006537795,0.00006828973,0.0002019974,0.00004427985,0.0001051031,0.0005375447],"genre_scores_gemma":[0.992371,0.000007242393,0.006945692,0.0001266497,0.00007984841,0.00002661187,0.0004216708,0.000008524658,0.00001272926],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8927625,"threshold_uncertainty_score":0.4565502,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07721719914288039,"score_gpt":0.3531169046251268,"score_spread":0.2758997054822465,"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."}}