{"id":"W7126429335","doi":"10.18653/v1/2024.findings-eacl.14","title":"An Examination of the Robustness of Reference-Free Image Captioning Evaluation Metrics","year":2024,"lang":"","type":"article","venue":"","topic":"Multimodal Machine Learning Applications","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Canadian Institute for Advanced Research; Nvidia","keywords":"Closed captioning; Robustness (evolution); Image (mathematics); Image processing; Pattern recognition (psychology)","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"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":[0.004429057,0.0002072255,0.0002412697,0.0005367206,0.0002182925,0.0003060761,0.002168872,0.0001498568,0.0003163182],"category_scores_gemma":[0.001169728,0.0001617099,0.0001139421,0.004044233,0.0002095404,0.001052888,0.0005012456,0.0004270756,0.00001519711],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001854409,"about_ca_system_score_gemma":0.0004879581,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000884264,"about_ca_topic_score_gemma":0.00006442217,"domain_scores_codex":[0.9958652,0.001024673,0.000772235,0.0006561512,0.001458836,0.0002229408],"domain_scores_gemma":[0.9954562,0.0004979314,0.0004326111,0.002085597,0.001456993,0.0000707093],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00000299458,0.0002363919,0.0005008716,0.0002937547,0.00004344192,2.219985e-7,0.002550188,0.1989894,0.01579428,0.1217566,0.00007357453,0.6597583],"study_design_scores_gemma":[0.0001913708,0.00008405143,0.1637678,0.0001466271,0.0001104045,0.000003308273,0.0001325983,0.8298213,0.004279813,0.001301266,0.00002505985,0.0001363802],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3086849,0.0003319808,0.6828663,0.000741981,0.0004992457,0.0006304924,0.00001627815,0.0000857576,0.006143033],"genre_scores_gemma":[0.9339515,0.00002164584,0.06566802,0.000009124821,0.00007593081,0.00005399736,0.00001496817,0.00001817556,0.0001866164],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6596219,"threshold_uncertainty_score":0.6594337,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05109654786901989,"score_gpt":0.3427500987957064,"score_spread":0.2916535509266865,"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."}}