{"id":"W1511840314","doi":"10.1109/icra.2015.7139526","title":"Tracking benchmark and evaluation for manipulation tasks","year":2015,"lang":"en","type":"article","venue":"","topic":"Video Surveillance and Tracking Methods","field":"Computer Science","cited_by":32,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"BitTorrent tracker; Computer science; Ground truth; Benchmark (surveying); Metric (unit); Artificial intelligence; Tracking (education); Computer vision; Process (computing); Video tracking; Scripting language; Face (sociological concept); Sensitivity (control systems); Object (grammar); Eye tracking; Engineering","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006052153,0.003648848,0.00213774,0.004171359,0.001786614,0.002502448,0.00341775,0.003176546,0.006644618],"category_scores_gemma":[0.01799548,0.0006178216,0.001608669,0.003904592,0.0009112122,0.00228659,0.002358679,0.001959738,0.007974257],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002322035,"about_ca_system_score_gemma":0.00234688,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02319656,"about_ca_topic_score_gemma":0.02893924,"domain_scores_codex":[0.9914002,0.001546342,0.001080828,0.002240957,0.003038179,0.0006935849],"domain_scores_gemma":[0.9912116,0.002077287,0.0007317798,0.002117777,0.00315823,0.000703374],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.002926984,0.002342498,0.01748919,0.004059857,0.001160688,0.0006362463,0.0003770827,0.07514523,0.0281558,0.003162177,0.3854209,0.4791233],"study_design_scores_gemma":[0.00130355,0.005321415,0.07389589,0.001063227,0.000678207,0.004284498,0.0007821257,0.6013774,0.07959235,0.008201209,0.2230052,0.0004950233],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2735645,0.02452933,0.3137704,0.001671265,0.006666051,0.006594695,0.1704802,0.1410197,0.06170394],"genre_scores_gemma":[0.230895,0.002813434,0.2022835,0.0008894732,0.0005492596,0.002362229,0.5389298,0.004123982,0.01715338],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02319656,"threshold_uncertainty_score":0.04612309,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1909700169973068,"score_gpt":0.3899478782084224,"score_spread":0.1989778612111156,"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."}}