{"id":"W4412889699","doi":"10.18653/v1/2025.acl-long.1470","title":"ViGiL3D: A Linguistically Diverse Dataset for 3D Visual Grounding","year":2025,"lang":"en","type":"article","venue":"","topic":"Human Pose and Action Recognition","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institute for Advanced Research","keywords":"Computer science; Ground; Artificial intelligence; Natural language processing; Engineering","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":[0.0001211294,0.00005846039,0.00006624731,0.00009103644,0.000168691,0.0001716301,0.0002184198,0.00002832209,0.0001292518],"category_scores_gemma":[0.0001069795,0.00005348837,0.00002905714,0.0001293331,0.00001389313,0.0001864131,0.0001243662,0.00004396367,0.0001389986],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000204809,"about_ca_system_score_gemma":0.0000291321,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001343996,"about_ca_topic_score_gemma":0.00001118231,"domain_scores_codex":[0.9994599,0.00001233678,0.0001120394,0.000206599,0.00007458151,0.0001345051],"domain_scores_gemma":[0.9996229,0.0001207293,0.00002170148,0.0001458611,0.00005203507,0.00003671691],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002121752,0.000183931,0.0001112398,0.00005630413,0.00004593901,0.00001355837,0.00008891847,0.000003418993,0.0004529891,0.5935801,0.1592019,0.2462405],"study_design_scores_gemma":[0.0008976349,0.0001359365,0.000357363,0.00005118543,0.00003164069,0.000004507698,0.00007437248,0.1274242,0.001775396,0.02543468,0.8435587,0.0002544378],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.000875838,0.000003803538,0.9873703,0.0002190626,0.0005076767,0.0001197114,0.00007096081,0.00009583763,0.01073684],"genre_scores_gemma":[0.5746817,0.00001221207,0.4072529,0.008355246,0.0006101938,0.00008854482,0.001443959,0.00001177243,0.007543447],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6843567,"threshold_uncertainty_score":0.2181192,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02781625247821911,"score_gpt":0.3431837223646392,"score_spread":0.3153674698864201,"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."}}