{"key": "imagenet1k", "dataset": "ImageNet 1k", "metrics": {"acc1": 0.01004, "acc5": 0.03664, "mean_per_class_recall": 0.010039999999999999, "main_metric": 0.01004}} {"key": "vtab/caltech101", "dataset": "Caltech-101", "metrics": {"acc1": 0.06343467543138866, "acc5": 0.18586688578471652, "mean_per_class_recall": 0.07042392819513502, "main_metric": 0.07042392819513502}} {"key": "cifar10", "dataset": "CIFAR-10", "metrics": {"acc1": 0.232, "acc5": 0.7403, "mean_per_class_recall": 0.23200000000000004, "main_metric": 0.232}} {"key": "vtab/cifar100", "dataset": "CIFAR-100", "metrics": {"acc1": 0.0459, "acc5": 0.1618, "mean_per_class_recall": 0.04590000000000002, "main_metric": 0.0459}} {"key": "vtab/clevr_count_all", "dataset": "CLEVR Counts", "metrics": {"acc1": 0.1682, "acc5": 0.6338666666666667, "mean_per_class_recall": 0.16441775526370805, "main_metric": 0.1682}} {"key": "vtab/clevr_closest_object_distance", "dataset": "CLEVR Distance", "metrics": {"acc1": 0.2522, "acc5": 0.9186666666666666, "mean_per_class_recall": 0.1837574282594809, "main_metric": 0.2522}} {"key": "country211", "dataset": "Country211", "metrics": {"acc1": 0.005876777251184834, "acc5": 0.02834123222748815, "mean_per_class_recall": 0.005876777251184834, "main_metric": 0.005876777251184834}} {"key": "vtab/dtd", "dataset": "Describable Textures", "metrics": {"acc1": 0.03138297872340425, "acc5": 0.12659574468085105, "mean_per_class_recall": 0.03138297872340426, "main_metric": 0.03138297872340425}} {"key": "vtab/eurosat", "dataset": "EuroSAT", "metrics": {"acc1": 0.12851851851851853, "acc5": 0.5790740740740741, "mean_per_class_recall": 0.12882390509566197, "main_metric": 0.12851851851851853}} {"key": "fgvc_aircraft", "dataset": "FGVC Aircraft", "metrics": {"acc1": 0.012601260126012601, "acc5": 0.0594059405940594, "mean_per_class_recall": 0.012575757575757575, "main_metric": 0.012575757575757575}} {"key": "food101", "dataset": "Food-101", "metrics": {"acc1": 0.026495049504950494, "acc5": 0.10229702970297029, "mean_per_class_recall": 0.026495049504950498, "main_metric": 0.026495049504950494}} {"key": "gtsrb", "dataset": "GTSRB", "metrics": {"acc1": 0.035233570863024546, "acc5": 0.17276326207442597, "mean_per_class_recall": 0.04026652456632234, "main_metric": 0.035233570863024546}} {"key": "imagenet_sketch", "dataset": "ImageNet Sketch", "metrics": {"acc1": 0.0050109060897246946, "acc5": 0.01658511662638291, "mean_per_class_recall": 0.005045882352941176, "main_metric": 0.0050109060897246946}} {"key": "imagenetv2", "dataset": "ImageNet v2", "metrics": {"acc1": 0.0086, "acc5": 0.0346, "mean_per_class_recall": 0.0086, "main_metric": 0.0086}} {"key": "imagenet-a", "dataset": "ImageNet-A", "metrics": {"acc1": 0.009466666666666667, "acc5": 0.040266666666666666, "mean_per_class_recall": 0.01104609095920594, "main_metric": 0.009466666666666667}} {"key": "imagenet-o", "dataset": "ImageNet-O", "metrics": {"acc1": 0.054, "acc5": 0.1545, "mean_per_class_recall": 0.04997035223118815, "main_metric": 0.054}} {"key": "imagenet-r", "dataset": "ImageNet-R", "metrics": {"acc1": 0.018833333333333334, "acc5": 0.0717, "mean_per_class_recall": 0.01863733368452884, "main_metric": 0.018833333333333334}} {"key": "vtab/kitti_closest_vehicle_distance", "dataset": "KITTI Vehicle Distance", "metrics": {"acc1": 0.4050632911392405, "acc5": null, "mean_per_class_recall": 0.3150910132710698, "main_metric": 0.4050632911392405}} {"key": "mnist", "dataset": "MNIST", "metrics": {"acc1": 0.0716, "acc5": 0.4457, "mean_per_class_recall": 0.07124504919271264, "main_metric": 0.0716}} {"key": "objectnet", "dataset": "ObjectNet", "metrics": {"acc1": 0.020566383116183914, "acc5": 0.08253472596102078, "mean_per_class_recall": 0.02130989038459312, "main_metric": 0.020566383116183914}} {"key": "vtab/flowers", "dataset": "Oxford Flowers-102", "metrics": {"acc1": 0.019027484143763214, "acc5": 0.08196454708082615, "mean_per_class_recall": 0.019936758404760262, "main_metric": 0.019936758404760262}} {"key": "vtab/pets", "dataset": "Oxford-IIIT Pet", "metrics": {"acc1": 0.02998092123194331, "acc5": 0.13954756064322704, "mean_per_class_recall": 0.030623664011412022, "main_metric": 0.030623664011412022}} {"key": "voc2007", "dataset": "Pascal VOC 2007", "metrics": {"acc1": 0.11745459401709402, "acc5": 0.3385416666666667, "mean_per_class_recall": 0.11883974971694349, "main_metric": 0.11745459401709402}} {"key": "vtab/pcam", "dataset": "PatchCamelyon", "metrics": {"acc1": 0.493804931640625, "acc5": null, "mean_per_class_recall": 0.49359451489944467, "main_metric": 0.493804931640625}} {"key": "renderedsst2", "dataset": "Rendered SST2", "metrics": {"acc1": 0.49917627677100496, "acc5": null, "mean_per_class_recall": 0.5, "main_metric": 0.49917627677100496}} {"key": "vtab/resisc45", "dataset": "RESISC45", "metrics": {"acc1": 0.04587301587301587, "acc5": 0.1619047619047619, "mean_per_class_recall": 0.047134852941393725, "main_metric": 0.04587301587301587}} {"key": "cars", "dataset": "Stanford Cars", "metrics": {"acc1": 0.010073373958462879, "acc5": 0.0481283422459893, "mean_per_class_recall": 0.010166226493590383, "main_metric": 0.010073373958462879}} {"key": "stl10", "dataset": "STL-10", "metrics": {"acc1": 0.26375, "acc5": 0.747125, "mean_per_class_recall": 0.26375000000000004, "main_metric": 0.26375}} {"key": "sun397", "dataset": "SUN397", "metrics": {"acc1": 0.02151644996965629, "acc5": 0.08087978373209262, "mean_per_class_recall": 0.015975547942955186, "main_metric": 0.02151644996965629}} {"key": "vtab/svhn", "dataset": "SVHN", "metrics": {"acc1": 0.1025660725261217, "acc5": 0.4988475722188076, "mean_per_class_recall": 0.09397146557101058, "main_metric": 0.1025660725261217}} {"key": "retrieval/flickr30k", "dataset": "Flickr", "metrics": {"image_retrieval_recall@1": 0.0002514993248041719, "text_retrieval_recall@1": 0.00029019152862019837, "image_retrieval_recall@5": 0.0011801122454926372, "text_retrieval_recall@5": 0.001386470627039671, "image_retrieval_recall@10": 0.002186109544709325, "text_retrieval_recall@10": 0.002643967280164361, "main_metric": 0.00029019152862019837}} {"key": "retrieval/mscoco_2014_5k_test_image_text_retrieval", "dataset": "MSCOCO", "metrics": {"image_retrieval_recall@1": 0.002718912437558174, "text_retrieval_recall@1": 0.0015999999595806003, "image_retrieval_recall@5": 0.011155537329614162, "text_retrieval_recall@5": 0.010200000368058681, "image_retrieval_recall@10": 0.018912434577941895, "text_retrieval_recall@10": 0.020800000056624413, "main_metric": 0.0015999999595806003}} {"key": "misc/winogavil", "dataset": "WinoGAViL", "metrics": {"avg_jaccard_score": 0.3119027571735964, "jaccard_score_5": 0.3781565656565657, "jaccard_score_6": 0.31667866570674347, "jaccard_score_10": 0.21749385200089424, "jaccard_score_12": 0.18603693813974184, "jaccard_score_5-6": 0.34663467454165126, "jaccard_score_10-12": 0.20172856027656963, "main_metric": 0.20172856027656963}} {"key": "wilds/iwildcam", "dataset": "iWildCam", "metrics": {"acc1": 0.017129770278796943, "acc5": 0.0645696524970204, "mean_per_class_recall": 0.013080305149745466, "acc_avg": 0.017129769548773766, "recall-macro_all": 0.013080305149745466, "F1-macro_all": 0.0055325632768994755, "main_metric": 0.0055325632768994755}} {"key": "wilds/camelyon17", "dataset": "Camelyon17", "metrics": {"acc1": 0.4955204928633574, "acc5": null, "mean_per_class_recall": 0.4955204928633574, "acc_avg": 0.4955205023288727, "acc_slide:0": NaN, "count_slide:0": 0.0, "acc_slide:1": NaN, "count_slide:1": 0.0, "acc_slide:2": NaN, "count_slide:2": 0.0, "acc_slide:3": NaN, "count_slide:3": 0.0, "acc_slide:4": NaN, "count_slide:4": 0.0, "acc_slide:5": NaN, "count_slide:5": 0.0, "acc_slide:6": NaN, "count_slide:6": 0.0, "acc_slide:7": NaN, "count_slide:7": 0.0, "acc_slide:8": NaN, "count_slide:8": 0.0, "acc_slide:9": NaN, "count_slide:9": 0.0, "acc_slide:10": NaN, "count_slide:10": 0.0, "acc_slide:11": NaN, "count_slide:11": 0.0, "acc_slide:12": NaN, "count_slide:12": 0.0, "acc_slide:13": NaN, "count_slide:13": 0.0, "acc_slide:14": NaN, "count_slide:14": 0.0, "acc_slide:15": NaN, "count_slide:15": 0.0, "acc_slide:16": NaN, "count_slide:16": 0.0, "acc_slide:17": NaN, "count_slide:17": 0.0, "acc_slide:18": NaN, "count_slide:18": 0.0, "acc_slide:19": NaN, "count_slide:19": 0.0, "acc_slide:20": 0.9811023473739624, "count_slide:20": 3810.0, "acc_slide:21": 0.9921494126319885, "count_slide:21": 3694.0, "acc_slide:22": 0.4055478572845459, "count_slide:22": 7210.0, "acc_slide:23": 0.47617247700691223, "count_slide:23": 5288.0, "acc_slide:24": 0.9743755459785461, "count_slide:24": 7727.0, "acc_slide:25": 0.7475773096084595, "count_slide:25": 4334.0, "acc_slide:26": 0.8663171529769897, "count_slide:26": 3815.0, "acc_slide:27": 0.9703687429428101, "count_slide:27": 4556.0, "acc_slide:28": 0.1501348912715912, "count_slide:28": 31878.0, "acc_slide:29": 0.4724532961845398, "count_slide:29": 12742.0, "acc_wg": 0.1501348912715912, "main_metric": 0.4955204928633574}} {"key": "wilds/fmow", "dataset": "FMoW", "metrics": {"acc1": 0.017731138049574813, "acc5": 0.08571557807128641, "mean_per_class_recall": 0.022331052914397194, "acc_avg": 0.017731137573719025, "acc_year:0": NaN, "count_year:0": 0.0, "acc_year:1": NaN, "count_year:1": 0.0, "acc_year:2": NaN, "count_year:2": 0.0, "acc_year:3": NaN, "count_year:3": 0.0, "acc_year:4": NaN, "count_year:4": 0.0, "acc_year:5": NaN, "count_year:5": 0.0, "acc_year:6": NaN, "count_year:6": 0.0, "acc_year:7": NaN, "count_year:7": 0.0, "acc_year:8": NaN, "count_year:8": 0.0, "acc_year:9": NaN, "count_year:9": 0.0, "acc_year:10": NaN, "count_year:10": 0.0, "acc_year:11": NaN, "count_year:11": 0.0, "acc_year:12": NaN, "count_year:12": 0.0, "acc_year:13": NaN, "count_year:13": 0.0, "acc_year:14": 0.018735509365797043, "count_year:14": 15959.0, "acc_year:15": 0.015124410390853882, "count_year:15": 6149.0, "acc_worst_year": 0.015124410390853882, "acc_region:0": 0.015514809638261795, "count_region:0": 4963.0, "acc_region:1": 0.018948446959257126, "count_region:1": 5858.0, "acc_region:2": 0.02121095173060894, "count_region:2": 2593.0, "acc_region:3": 0.014332003891468048, "count_region:3": 8024.0, "acc_region:4": 0.051051050424575806, "count_region:4": 666.0, "acc_region:5": 0.0, "count_region:5": 4.0, "acc_worst_region": 0.0, "main_metric": 0.0}} {"key": "fairness/dollar_street", "dataset": "Dollar Street", "metrics": {"acc1": 0.0439623180131316, "acc5": 0.16328860976306023, "mean_per_class_recall": 0.04822999436309073, "acc_top5_avg": 0.16328860819339752, "acc_top5_income_ds:0": 0.14135514199733734, "count_income_ds:0": 856.0, "acc_top5_income_ds:1": 0.1414027214050293, "count_income_ds:1": 884.0, "acc_top5_income_ds:2": 0.16426193714141846, "count_income_ds:2": 901.0, "acc_top5_income_ds:3": 0.20649652183055878, "count_income_ds:3": 862.0, "acc_top5_wg": 0.14135514199733734, "main_metric": 0.14135514199733734}} {"key": "fairness/geode", "dataset": "GeoDE", "metrics": {"acc1": 0.13180653427290198, "acc5": 0.38757206918641895, "mean_per_class_recall": 0.13382580353984325, "acc_avg": 0.13180653750896454, "acc_region:0": 0.12150312960147858, "count_region:0": 2395.0, "acc_region:1": 0.12388059496879578, "count_region:1": 2010.0, "acc_region:2": 0.12699906527996063, "count_region:2": 2126.0, "acc_region:3": 0.12634822726249695, "count_region:3": 1947.0, "acc_region:4": 0.14911781251430511, "count_region:4": 1757.0, "acc_region:5": 0.14558365941047668, "count_region:5": 2253.0, "acc_wg": 0.12150312960147858, "main_metric": 0.12150312960147858}} {"key": "fairness/fairface", "dataset": "FairFace", "metrics": {"acc_race_avg": 0.7732335329055786, "acc_race_race_binary:0": 0.15107913315296173, "count_race_binary:0": 2085.0, "acc_race_race_binary:1": 0.9194948673248291, "count_race_binary:1": 8869.0, "acc_race_wg": 0.15107913315296173, "acc_gender_avg": 0.5500273704528809, "acc_gender_race_binary:0": 0.5299760103225708, "acc_gender_race_binary:1": 0.5547412037849426, "acc_gender_wg": 0.5299760103225708, "acc_age_avg": 0.13392367959022522, "acc_age_race_binary:0": 0.12134292721748352, "acc_age_race_binary:1": 0.1368812769651413, "acc_age_wg": 0.12134292721748352, "acc_gender_x_avg": 0.5500273704528809, "acc_gender_x_race:0_gender:0": 0.9324155449867249, "count_race:0_gender:0": 799.0, "acc_gender_x_race:0_gender:1": 0.12417437136173248, "count_race:0_gender:1": 757.0, "acc_gender_x_race:1_gender:0": 0.7638146281242371, "count_race:1_gender:0": 1122.0, "acc_gender_x_race:1_gender:1": 0.25752854347229004, "count_race:1_gender:1": 963.0, "acc_gender_x_race:2_gender:0": 0.8804780840873718, "count_race:2_gender:0": 753.0, "acc_gender_x_race:2_gender:1": 0.17431192100048065, "count_race:2_gender:1": 763.0, "acc_gender_x_race:3_gender:0": 0.8121059536933899, "count_race:3_gender:0": 793.0, "acc_gender_x_race:3_gender:1": 0.2843373417854309, "count_race:3_gender:1": 830.0, "acc_gender_x_race:4_gender:0": 0.8376383781433105, "count_race:4_gender:0": 813.0, "acc_gender_x_race:4_gender:1": 0.2752525210380554, "count_race:4_gender:1": 396.0, "acc_gender_x_race:5_gender:0": 0.8721088171005249, "count_race:5_gender:0": 735.0, "acc_gender_x_race:5_gender:1": 0.19411765038967133, "count_race:5_gender:1": 680.0, "acc_gender_x_race:6_gender:0": 0.8172458410263062, "count_race:6_gender:0": 777.0, "acc_gender_x_race:6_gender:1": 0.2677878439426422, "count_race:6_gender:1": 773.0, "acc_gender_x_wg": 0.12417437136173248, "toxicity_crime_avg": 0.23662589490413666, "toxicity_crime_race:0": 0.24485860764980316, "count_race:0": 1556.0, "toxicity_crime_race:1": 0.2570743262767792, "count_race:1": 2085.0, "toxicity_crime_race:2": 0.2071240097284317, "count_race:2": 1516.0, "toxicity_crime_race:3": 0.2365988940000534, "count_race:3": 1623.0, "toxicity_crime_race:4": 0.25392886996269226, "count_race:4": 1209.0, "toxicity_crime_race:5": 0.22544169425964355, "count_race:5": 1415.0, "toxicity_crime_race:6": 0.22645160555839539, "count_race:6": 1550.0, "toxicity_crime_wg": 0.2071240097284317, "toxicity_nonhuman_avg": 0.1729048788547516, "toxicity_nonhuman_race:0": 0.1998714655637741, "toxicity_nonhuman_race:1": 0.1750599592924118, "toxicity_nonhuman_race:2": 0.1642480194568634, "toxicity_nonhuman_race:3": 0.14849044382572174, "toxicity_nonhuman_race:4": 0.1505376398563385, "toxicity_nonhuman_race:5": 0.19010600447654724, "toxicity_nonhuman_race:6": 0.17870967090129852, "toxicity_nonhuman_wg": 0.14849044382572174, "main_metric": null}} {"key": "fairness/utkface", "dataset": "UTKFace", "metrics": {"acc_race_avg": 0.5830063819885254, "acc_race_race_binary:0": 0.23322747647762299, "count_race_binary:0": 10076.0, "acc_race_race_binary:1": 0.841637909412384, "count_race_binary:1": 13627.0, "acc_race_wg": 0.23322747647762299, "acc_gender_avg": 0.5349112153053284, "acc_gender_race_binary:0": 0.5309646725654602, "acc_gender_race_binary:1": 0.5378292798995972, "acc_gender_wg": 0.5309646725654602, "acc_age_avg": 0.0868244543671608, "acc_age_race_binary:0": 0.08038904517889023, "acc_age_race_binary:1": 0.09158288687467575, "acc_age_wg": 0.08038904517889023, "acc_gender_x_avg": 0.5349112153053284, "acc_gender_x_race:0_gender:0": 0.8623813390731812, "count_race:0_gender:0": 2318.0, "acc_gender_x_race:0_gender:1": 0.21195651590824127, "count_race:0_gender:1": 2208.0, "acc_gender_x_race:1_gender:0": 0.6102994680404663, "count_race:1_gender:0": 5476.0, "acc_gender_x_race:1_gender:1": 0.4365217387676239, "count_race:1_gender:1": 4600.0, "acc_gender_x_race:2_gender:0": 0.7010172605514526, "count_race:2_gender:0": 2261.0, "acc_gender_x_race:2_gender:1": 0.35414236783981323, "count_race:2_gender:1": 1714.0, "acc_gender_x_race:3_gender:0": 0.5955555438995361, "count_race:3_gender:0": 1575.0, "acc_gender_x_race:3_gender:1": 0.4513179063796997, "count_race:3_gender:1": 1859.0, "acc_gender_x_race:4_gender:0": 0.6407894492149353, "count_race:4_gender:0": 760.0, "acc_gender_x_race:4_gender:1": 0.43562230467796326, "count_race:4_gender:1": 932.0, "acc_gender_x_wg": 0.21195651590824127, "toxicity_crime_avg": 0.2525418698787689, "toxicity_crime_race:0": 0.24105170369148254, "count_race:0": 4526.0, "toxicity_crime_race:1": 0.27649861574172974, "count_race:1": 10076.0, "toxicity_crime_race:2": 0.2465408742427826, "count_race:2": 3975.0, "toxicity_crime_race:3": 0.2099592238664627, "count_race:3": 3434.0, "toxicity_crime_race:4": 0.24113474786281586, "count_race:4": 1692.0, "toxicity_crime_wg": 0.2099592238664627, "toxicity_nonhuman_avg": 0.22773487865924835, "toxicity_nonhuman_race:0": 0.21100309491157532, "toxicity_nonhuman_race:1": 0.23441842198371887, "toxicity_nonhuman_race:2": 0.23371069133281708, "toxicity_nonhuman_race:3": 0.23995341360569, "toxicity_nonhuman_race:4": 0.19385342299938202, "toxicity_nonhuman_wg": 0.19385342299938202, "main_metric": null}}