{"id":"W7131658085","doi":"10.5220/0014376400004918","title":"Evaluating Vision-Based Object Recognition Algorithms for Autonomous Vehicles with Varying Environmental Conditions: A Benchmark Study","year":2025,"lang":"","type":"article","venue":"","topic":"Autonomous Vehicle Technology and Safety","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Benchmark (surveying); Object (grammar); Cognitive neuroscience of visual object recognition; Noise (video); Pattern recognition (psychology); Feature (linguistics)","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":[0.002444191,0.001598972,0.001060555,0.001485535,0.0006738671,0.001487619,0.001727419,0.001937763,0.0009285755],"category_scores_gemma":[0.008283748,0.0003728315,0.001035569,0.00118972,0.0007454089,0.00145807,0.001025136,0.0006979203,0.0005373002],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001267963,"about_ca_system_score_gemma":0.001127732,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01613612,"about_ca_topic_score_gemma":0.01443955,"domain_scores_codex":[0.9973763,0.0005331612,0.0002472549,0.0007019783,0.0009009761,0.0002402193],"domain_scores_gemma":[0.9943988,0.002176211,0.0004788311,0.0006564339,0.002054547,0.0002351912],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.004776782,0.005090883,0.03745072,0.001212157,0.001268407,0.0003385899,0.0002767757,0.5491737,0.07351167,0.0009657274,0.004986735,0.3209479],"study_design_scores_gemma":[0.0002873806,0.005202128,0.03839583,0.00003859385,0.0002700587,0.0003659236,0.0004444475,0.9108552,0.04159538,0.0007939979,0.001674305,0.00007681727],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9674802,0.001105515,0.02799105,0.00009582087,0.0001074515,0.0002833096,0.0008365651,0.0005128767,0.001587083],"genre_scores_gemma":[0.9644333,0.000462443,0.02961581,0.00005792492,0.0000398272,0.0001089747,0.003885719,0.0001028603,0.001293195],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01613612,"threshold_uncertainty_score":0.03208435,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02449752033413881,"score_gpt":0.3081456771123923,"score_spread":0.2836481567782534,"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."}}