{"id":"W4385612782","doi":"10.1145/3539618.3591903","title":"AToMiC: An Image/Text Retrieval Test Collection to Support Multimedia Content Creation","year":2023,"lang":"en","type":"article","venue":"","topic":"Multimodal Machine Learning Applications","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Test (biology); Multimedia; Content (measure theory); Information retrieval; Image retrieval; Content-based image retrieval; Image (mathematics); Artificial intelligence","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":[0.002723806,0.002727804,0.0013482,0.006268806,0.001692671,0.002179368,0.003875521,0.002693864,0.01209863],"category_scores_gemma":[0.009756105,0.00061316,0.00175166,0.004487487,0.001225282,0.003661848,0.003578406,0.002306556,0.01819993],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00170186,"about_ca_system_score_gemma":0.001825271,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01536094,"about_ca_topic_score_gemma":0.03332415,"domain_scores_codex":[0.9970545,0.0006231572,0.0003398018,0.0006690689,0.001031385,0.0002821128],"domain_scores_gemma":[0.9940627,0.001368796,0.000336648,0.002029215,0.001503896,0.0006986707],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001091842,0.001773913,0.006210141,0.002586188,0.000309649,0.0006265154,0.0004157721,0.005611339,0.01588014,0.002315505,0.8274655,0.1357135],"study_design_scores_gemma":[0.001748517,0.002482222,0.03730281,0.0005033955,0.0004894325,0.004402496,0.001984046,0.1402342,0.07757399,0.00688846,0.7258204,0.000570095],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"dataset","genre_gemma":"methods","genre_scores_codex":[0.1411764,0.004065787,0.05748933,0.001725541,0.001885073,0.005970068,0.6890669,0.0740936,0.02452738],"genre_scores_gemma":[0.04823869,0.0003903095,0.05312537,0.0004665499,0.000223198,0.001545178,0.8878052,0.001517344,0.006688157],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.01536094,"threshold_uncertainty_score":0.040474,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03756938073927824,"score_gpt":0.3233776406108255,"score_spread":0.2858082598715473,"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."}}