50 percent experiment

breaking_0.5_trained/50_most_difficult/eval_results.jsonl

@@ -0,0 +1,40 @@
+{"key": "vtab/caltech101", "dataset": "Caltech-101", "metrics": {"acc1": 0.09268693508627773, "acc5": 0.4236647493837305, "mean_per_class_recall": 0.14870367510668095, "main_metric": 0.14870367510668095}}
+{"key": "cifar10", "dataset": "CIFAR-10", "metrics": {"acc1": 0.3651, "acc5": 0.9133, "mean_per_class_recall": 0.36510000000000004, "main_metric": 0.3651}}
+{"key": "vtab/cifar100", "dataset": "CIFAR-100", "metrics": {"acc1": 0.1107, "acc5": 0.3287, "mean_per_class_recall": 0.11069999999999998, "main_metric": 0.1107}}
+{"key": "vtab/clevr_count_all", "dataset": "CLEVR Counts", "metrics": {"acc1": 0.1422, "acc5": 0.6363333333333333, "mean_per_class_recall": 0.1425002016439411, "main_metric": 0.1422}}
+{"key": "vtab/clevr_closest_object_distance", "dataset": "CLEVR Distance", "metrics": {"acc1": 0.21933333333333332, "acc5": 0.9186666666666666, "mean_per_class_recall": 0.25125632339064236, "main_metric": 0.21933333333333332}}
+{"key": "country211", "dataset": "Country211", "metrics": {"acc1": 0.011563981042654029, "acc5": 0.05085308056872038, "mean_per_class_recall": 0.01156398104265403, "main_metric": 0.011563981042654029}}
+{"key": "vtab/dtd", "dataset": "Describable Textures", "metrics": {"acc1": 0.05319148936170213, "acc5": 0.16170212765957448, "mean_per_class_recall": 0.05319148936170212, "main_metric": 0.05319148936170213}}
+{"key": "vtab/eurosat", "dataset": "EuroSAT", "metrics": {"acc1": 0.16777777777777778, "acc5": 0.5309259259259259, "mean_per_class_recall": 0.1596294452891564, "main_metric": 0.16777777777777778}}
+{"key": "fgvc_aircraft", "dataset": "FGVC Aircraft", "metrics": {"acc1": 0.010801080108010801, "acc5": 0.050405040504050404, "mean_per_class_recall": 0.010597147950089125, "main_metric": 0.010597147950089125}}
+{"key": "food101", "dataset": "Food-101", "metrics": {"acc1": 0.04388118811881188, "acc5": 0.16324752475247525, "mean_per_class_recall": 0.04388118811881188, "main_metric": 0.04388118811881188}}
+{"key": "gtsrb", "dataset": "GTSRB", "metrics": {"acc1": 0.04069675376088678, "acc5": 0.19501187648456056, "mean_per_class_recall": 0.055558123157971484, "main_metric": 0.04069675376088678}}
+{"key": "imagenet1k", "dataset": "ImageNet 1k", "metrics": {"acc1": 0.02696, "acc5": 0.09326, "mean_per_class_recall": 0.026920000000000003, "main_metric": 0.02696}}
+{"key": "imagenet_sketch", "dataset": "ImageNet Sketch", "metrics": {"acc1": 0.010591679930829845, "acc5": 0.03075320796242803, "mean_per_class_recall": 0.010595686274509804, "main_metric": 0.010591679930829845}}
+{"key": "imagenetv2", "dataset": "ImageNet v2", "metrics": {"acc1": 0.0249, "acc5": 0.0925, "mean_per_class_recall": 0.024900000000000002, "main_metric": 0.0249}}
+{"key": "imagenet-a", "dataset": "ImageNet-A", "metrics": {"acc1": 0.015866666666666668, "acc5": 0.06866666666666667, "mean_per_class_recall": 0.02080801451118908, "main_metric": 0.015866666666666668}}
+{"key": "imagenet-o", "dataset": "ImageNet-O", "metrics": {"acc1": 0.098, "acc5": 0.259, "mean_per_class_recall": 0.08780323104827747, "main_metric": 0.098}}
+{"key": "imagenet-r", "dataset": "ImageNet-R", "metrics": {"acc1": 0.042, "acc5": 0.13756666666666667, "mean_per_class_recall": 0.03867133926402023, "main_metric": 0.042}}
+{"key": "vtab/kitti_closest_vehicle_distance", "dataset": "KITTI Vehicle Distance", "metrics": {"acc1": 0.30520393811533053, "acc5": null, "mean_per_class_recall": 0.27514037740141123, "main_metric": 0.30520393811533053}}
+{"key": "mnist", "dataset": "MNIST", "metrics": {"acc1": 0.101, "acc5": 0.5316, "mean_per_class_recall": 0.09828196632892862, "main_metric": 0.101}}
+{"key": "objectnet", "dataset": "ObjectNet", "metrics": {"acc1": 0.03564121890815118, "acc5": 0.13325078066113924, "mean_per_class_recall": 0.03625983462133988, "main_metric": 0.03564121890815118}}
+{"key": "vtab/flowers", "dataset": "Oxford Flowers-102", "metrics": {"acc1": 0.03805496828752643, "acc5": 0.12457310131728737, "mean_per_class_recall": 0.04278602525010276, "main_metric": 0.04278602525010276}}
+{"key": "vtab/pets", "dataset": "Oxford-IIIT Pet", "metrics": {"acc1": 0.05423821204687926, "acc5": 0.21068411011174706, "mean_per_class_recall": 0.053849807001500674, "main_metric": 0.053849807001500674}}
+{"key": "voc2007", "dataset": "Pascal VOC 2007", "metrics": {"acc1": 0.25834668803418803, "acc5": 0.6329460470085471, "mean_per_class_recall": 0.2915637982199252, "main_metric": 0.25834668803418803}}
+{"key": "vtab/pcam", "dataset": "PatchCamelyon", "metrics": {"acc1": 0.527740478515625, "acc5": null, "mean_per_class_recall": 0.5279375999008954, "main_metric": 0.527740478515625}}
+{"key": "renderedsst2", "dataset": "Rendered SST2", "metrics": {"acc1": 0.4986271279516749, "acc5": null, "mean_per_class_recall": 0.49944994499449946, "main_metric": 0.4986271279516749}}
+{"key": "vtab/resisc45", "dataset": "RESISC45", "metrics": {"acc1": 0.07380952380952381, "acc5": 0.25047619047619046, "mean_per_class_recall": 0.07417600471262596, "main_metric": 0.07380952380952381}}
+{"key": "cars", "dataset": "Stanford Cars", "metrics": {"acc1": 0.015545330182813084, "acc5": 0.061435144882477304, "mean_per_class_recall": 0.015701529135830537, "main_metric": 0.015545330182813084}}
+{"key": "stl10", "dataset": "STL-10", "metrics": {"acc1": 0.45475, "acc5": 0.944625, "mean_per_class_recall": 0.45474999999999993, "main_metric": 0.45475}}
+{"key": "sun397", "dataset": "SUN397", "metrics": {"acc1": 0.066940066572264, "acc5": 0.1955882082498115, "mean_per_class_recall": 0.05110946147092936, "main_metric": 0.066940066572264}}
+{"key": "vtab/svhn", "dataset": "SVHN", "metrics": {"acc1": 0.08577904118008604, "acc5": 0.5545482483097726, "mean_per_class_recall": 0.09885579279028249, "main_metric": 0.08577904118008604}}
+{"key": "retrieval/flickr_1k_test_image_text_retrieval", "dataset": "Flickr", "metrics": {"image_retrieval_recall@1": 0.022199999541044235, "text_retrieval_recall@1": 0.03200000151991844, "image_retrieval_recall@5": 0.066600002348423, "text_retrieval_recall@5": 0.0989999994635582, "image_retrieval_recall@10": 0.10859999805688858, "text_retrieval_recall@10": 0.15399999916553497, "mean_recall@1": 0.02710000053048134, "main_metric": 0.02710000053048134}}
+{"key": "retrieval/mscoco_2014_5k_test_image_text_retrieval", "dataset": "MSCOCO", "metrics": {"image_retrieval_recall@1": 0.01043582521378994, "text_retrieval_recall@1": 0.014800000004470348, "image_retrieval_recall@5": 0.03598560392856598, "text_retrieval_recall@5": 0.05400000140070915, "image_retrieval_recall@10": 0.0608956404030323, "text_retrieval_recall@10": 0.08839999884366989, "mean_recall@1": 0.012617912609130144, "main_metric": 0.012617912609130144}}
+{"key": "misc/winogavil", "dataset": "WinoGAViL", "metrics": {"avg_jaccard_score": 0.3861152319473959, "jaccard_score_5": 0.45285353535353534, "jaccard_score_6": 0.3820494360451164, "jaccard_score_10": 0.31220657276995306, "jaccard_score_12": 0.2670449488206497, "jaccard_score_5-6": 0.41654977236372587, "jaccard_score_10-12": 0.28957287833166057, "main_metric": 0.28957287833166057}}
+{"key": "wilds/iwildcam", "dataset": "iWildCam", "metrics": {"acc1": 0.0012152088055899606, "acc5": 0.04648173681381599, "mean_per_class_recall": 0.005373134975732705, "acc_avg": 0.001215208787471056, "recall-macro_all": 0.005373134975732705, "F1-macro_all": 0.001915663192964824, "main_metric": 0.001915663192964824}}
+{"key": "wilds/camelyon17", "dataset": "Camelyon17", "metrics": {"acc1": 0.5217508876713617, "acc5": null, "mean_per_class_recall": 0.5217508876713617, "acc_avg": 0.5217508673667908, "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": 0.09160105139017105, "count_slide:20": 3810.0, "acc_slide:21": 0.03302653133869171, "count_slide:21": 3694.0, "acc_slide:22": 0.5943134427070618, "count_slide:22": 7210.0, "acc_slide:23": 0.530635416507721, "count_slide:23": 5288.0, "acc_slide:24": 0.04658988118171692, "count_slide:24": 7727.0, "acc_slide:25": 0.2609598636627197, "count_slide:25": 4334.0, "acc_slide:26": 0.16566185653209686, "count_slide:26": 3815.0, "acc_slide:27": 0.05158033221960068, "count_slide:27": 4556.0, "acc_slide:28": 0.851119875907898, "count_slide:28": 31878.0, "acc_slide:29": 0.5748705267906189, "count_slide:29": 12742.0, "acc_wg": 0.03302653133869171, "main_metric": 0.5217508876713617}}
+{"key": "wilds/fmow", "dataset": "FMoW", "metrics": {"acc1": 0.012846028586936856, "acc5": 0.0759453591460105, "mean_per_class_recall": 0.02082277582669382, "acc_avg": 0.012846028432250023, "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": 0.012908076867461205, "count_year:14": 15959.0, "acc_year:15": 0.012684989720582962, "count_year:15": 6149.0, "acc_worst_year": 0.012684989720582962, "acc_region:0": 0.00947007816284895, "count_region:0": 4963.0, "acc_region:1": 0.016729259863495827, "count_region:1": 5858.0, "acc_region:2": 0.0053991517052054405, "count_region:2": 2593.0, "acc_region:3": 0.013958125375211239, "count_region:3": 8024.0, "acc_region:4": 0.019519519060850143, "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": 0.09192121039109336, "acc5": 0.28889523265772193, "mean_per_class_recall": 0.0953918917796648, "acc_top5_avg": 0.28889521956443787, "acc_top5_income_ds:0": 0.22079439461231232, "count_income_ds:0": 856.0, "acc_top5_income_ds:1": 0.25678732991218567, "count_income_ds:1": 884.0, "acc_top5_income_ds:2": 0.31076580286026, "count_income_ds:2": 901.0, "acc_top5_income_ds:3": 0.36658933758735657, "count_income_ds:3": 862.0, "acc_top5_wg": 0.22079439461231232, "main_metric": 0.22079439461231232}}
+{"key": "fairness/geode", "dataset": "GeoDE", "metrics": {"acc1": 0.24103139013452915, "acc5": 0.585281870595772, "mean_per_class_recall": 0.24172843685021853, "acc_avg": 0.24103139340877533, "acc_region:0": 0.2121085524559021, "count_region:0": 2395.0, "acc_region:1": 0.2457711398601532, "count_region:1": 2010.0, "acc_region:2": 0.24317967891693115, "count_region:2": 2126.0, "acc_region:3": 0.22239342331886292, "count_region:3": 1947.0, "acc_region:4": 0.2595332860946655, "count_region:4": 1757.0, "acc_region:5": 0.2671992778778076, "count_region:5": 2253.0, "acc_wg": 0.2121085524559021, "main_metric": 0.2121085524559021}}
+{"key": "fairness/fairface", "dataset": "FairFace", "metrics": {"acc_race_avg": 0.721745491027832, "acc_race_race_binary:0": 0.24076738953590393, "count_race_binary:0": 2085.0, "acc_race_race_binary:1": 0.8348178863525391, "count_race_binary:1": 8869.0, "acc_race_wg": 0.24076738953590393, "acc_gender_avg": 0.5227314233779907, "acc_gender_race_binary:0": 0.4954436421394348, "acc_gender_race_binary:1": 0.5291464924812317, "acc_gender_wg": 0.4954436421394348, "acc_age_avg": 0.05459193140268326, "acc_age_race_binary:0": 0.052278175950050354, "acc_age_race_binary:1": 0.05513586476445198, "acc_age_wg": 0.052278175950050354, "acc_gender_x_avg": 0.5227314233779907, "acc_gender_x_race:0_gender:0": 0.3679599463939667, "count_race:0_gender:0": 799.0, "acc_gender_x_race:0_gender:1": 0.733157217502594, "count_race:0_gender:1": 757.0, "acc_gender_x_race:1_gender:0": 0.1506238877773285, "count_race:1_gender:0": 1122.0, "acc_gender_x_race:1_gender:1": 0.8971962332725525, "count_race:1_gender:1": 963.0, "acc_gender_x_race:2_gender:0": 0.3200531303882599, "count_race:2_gender:0": 753.0, "acc_gender_x_race:2_gender:1": 0.7942332625389099, "count_race:2_gender:1": 763.0, "acc_gender_x_race:3_gender:0": 0.18915510177612305, "count_race:3_gender:0": 793.0, "acc_gender_x_race:3_gender:1": 0.9036144614219666, "count_race:3_gender:1": 830.0, "acc_gender_x_race:4_gender:0": 0.2263222634792328, "count_race:4_gender:0": 813.0, "acc_gender_x_race:4_gender:1": 0.8484848737716675, "count_race:4_gender:1": 396.0, "acc_gender_x_race:5_gender:0": 0.21360544860363007, "count_race:5_gender:0": 735.0, "acc_gender_x_race:5_gender:1": 0.8720588088035583, "count_race:5_gender:1": 680.0, "acc_gender_x_race:6_gender:0": 0.17245817184448242, "count_race:6_gender:0": 777.0, "acc_gender_x_race:6_gender:1": 0.8965071439743042, "count_race:6_gender:1": 773.0, "acc_gender_x_wg": 0.1506238877773285, "toxicity_crime_avg": 0.015793317928910255, "toxicity_crime_race:0": 0.010282776318490505, "count_race:0": 1556.0, "toxicity_crime_race:1": 0.02158273383975029, "count_race:1": 2085.0, "toxicity_crime_race:2": 0.009894459508359432, "count_race:2": 1516.0, "toxicity_crime_race:3": 0.02218114584684372, "count_race:3": 1623.0, "toxicity_crime_race:4": 0.012406948022544384, "count_race:4": 1209.0, "toxicity_crime_race:5": 0.012014133855700493, "count_race:5": 1415.0, "toxicity_crime_race:6": 0.018709678202867508, "count_race:6": 1550.0, "toxicity_crime_wg": 0.009894459508359432, "toxicity_nonhuman_avg": 0.23406974971294403, "toxicity_nonhuman_race:0": 0.38046273589134216, "toxicity_nonhuman_race:1": 0.16354915499687195, "toxicity_nonhuman_race:2": 0.30804750323295593, "toxicity_nonhuman_race:3": 0.19654959440231323, "toxicity_nonhuman_race:4": 0.20264682173728943, "toxicity_nonhuman_race:5": 0.22897526621818542, "toxicity_nonhuman_race:6": 0.1780645102262497, "toxicity_nonhuman_wg": 0.16354915499687195, "main_metric": null}}
+{"key": "fairness/utkface", "dataset": "UTKFace", "metrics": {"acc_race_avg": 0.5844407677650452, "acc_race_race_binary:0": 0.46347758173942566, "count_race_binary:0": 10076.0, "acc_race_race_binary:1": 0.6738827228546143, "count_race_binary:1": 13627.0, "acc_race_wg": 0.46347758173942566, "acc_gender_avg": 0.505252480506897, "acc_gender_race_binary:0": 0.47350138425827026, "acc_gender_race_binary:1": 0.5287297368049622, "acc_gender_wg": 0.47350138425827026, "acc_age_avg": 0.06336750835180283, "acc_age_race_binary:0": 0.08128225803375244, "acc_age_race_binary:1": 0.05012108385562897, "acc_age_wg": 0.05012108385562897, "acc_gender_x_avg": 0.505252480506897, "acc_gender_x_race:0_gender:0": 0.1902502179145813, "count_race:0_gender:0": 2318.0, "acc_gender_x_race:0_gender:1": 0.9270833134651184, "count_race:0_gender:1": 2208.0, "acc_gender_x_race:1_gender:0": 0.06482834368944168, "count_race:1_gender:0": 5476.0, "acc_gender_x_race:1_gender:1": 0.9599999785423279, "count_race:1_gender:1": 4600.0, "acc_gender_x_race:2_gender:0": 0.09287925809621811, "count_race:2_gender:0": 2261.0, "acc_gender_x_race:2_gender:1": 0.9702450633049011, "count_race:2_gender:1": 1714.0, "acc_gender_x_race:3_gender:0": 0.04380952566862106, "count_race:3_gender:0": 1575.0, "acc_gender_x_race:3_gender:1": 0.9774072170257568, "count_race:3_gender:1": 1859.0, "acc_gender_x_race:4_gender:0": 0.057894736528396606, "count_race:4_gender:0": 760.0, "acc_gender_x_race:4_gender:1": 0.9806867241859436, "count_race:4_gender:1": 932.0, "acc_gender_x_wg": 0.04380952566862106, "toxicity_crime_avg": 0.016073914244771004, "toxicity_crime_race:0": 0.008174989372491837, "count_race:0": 4526.0, "toxicity_crime_race:1": 0.025605399161577225, "count_race:1": 10076.0, "toxicity_crime_race:2": 0.0060377358458936214, "count_race:2": 3975.0, "toxicity_crime_race:3": 0.011648223735392094, "count_race:3": 3434.0, "toxicity_crime_race:4": 0.013002363964915276, "count_race:4": 1692.0, "toxicity_crime_wg": 0.0060377358458936214, "toxicity_nonhuman_avg": 0.1716238409280777, "toxicity_nonhuman_race:0": 0.310649573802948, "toxicity_nonhuman_race:1": 0.11184994131326675, "toxicity_nonhuman_race:2": 0.23094339668750763, "toxicity_nonhuman_race:3": 0.11531741172075272, "toxicity_nonhuman_race:4": 0.13061465322971344, "toxicity_nonhuman_wg": 0.11184994131326675, "main_metric": null}}