added evals
Browse files
breaking_0.1_trained/10_most_difficult/eval_results.jsonl
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{"key": "vtab/caltech101", "dataset": "Caltech-101", "metrics": {"acc1": 0.024815119145439605, "acc5": 0.11339359079704191, "mean_per_class_recall": 0.030406278197525994, "main_metric": 0.030406278197525994}}
|
| 2 |
+
{"key": "cifar10", "dataset": "CIFAR-10", "metrics": {"acc1": 0.1815, "acc5": 0.6719, "mean_per_class_recall": 0.1815, "main_metric": 0.1815}}
|
| 3 |
+
{"key": "vtab/cifar100", "dataset": "CIFAR-100", "metrics": {"acc1": 0.0432, "acc5": 0.1485, "mean_per_class_recall": 0.043199999999999995, "main_metric": 0.0432}}
|
| 4 |
+
{"key": "vtab/clevr_count_all", "dataset": "CLEVR Counts", "metrics": {"acc1": 0.13486666666666666, "acc5": 0.6739333333333334, "mean_per_class_recall": 0.13522116957063923, "main_metric": 0.13486666666666666}}
|
| 5 |
+
{"key": "vtab/clevr_closest_object_distance", "dataset": "CLEVR Distance", "metrics": {"acc1": 0.21513333333333334, "acc5": 0.9186666666666666, "mean_per_class_recall": 0.170386471126133, "main_metric": 0.21513333333333334}}
|
| 6 |
+
{"key": "country211", "dataset": "Country211", "metrics": {"acc1": 0.007203791469194313, "acc5": 0.02834123222748815, "mean_per_class_recall": 0.007203791469194313, "main_metric": 0.007203791469194313}}
|
| 7 |
+
{"key": "vtab/dtd", "dataset": "Describable Textures", "metrics": {"acc1": 0.027127659574468086, "acc5": 0.10691489361702128, "mean_per_class_recall": 0.027127659574468083, "main_metric": 0.027127659574468086}}
|
| 8 |
+
{"key": "vtab/eurosat", "dataset": "EuroSAT", "metrics": {"acc1": 0.15888888888888889, "acc5": 0.5357407407407407, "mean_per_class_recall": 0.14321307023392, "main_metric": 0.15888888888888889}}
|
| 9 |
+
{"key": "fgvc_aircraft", "dataset": "FGVC Aircraft", "metrics": {"acc1": 0.011701170117011701, "acc5": 0.051005100510051006, "mean_per_class_recall": 0.011702317290552584, "main_metric": 0.011702317290552584}}
|
| 10 |
+
{"key": "food101", "dataset": "Food-101", "metrics": {"acc1": 0.013544554455445544, "acc5": 0.06463366336633664, "mean_per_class_recall": 0.013544554455445548, "main_metric": 0.013544554455445544}}
|
| 11 |
+
{"key": "gtsrb", "dataset": "GTSRB", "metrics": {"acc1": 0.04338875692794933, "acc5": 0.2163895486935867, "mean_per_class_recall": 0.03357316853651535, "main_metric": 0.04338875692794933}}
|
| 12 |
+
{"key": "imagenet1k", "dataset": "ImageNet 1k", "metrics": {"acc1": 0.00562, "acc5": 0.02382, "mean_per_class_recall": 0.00562, "main_metric": 0.00562}}
|
| 13 |
+
{"key": "imagenet_sketch", "dataset": "ImageNet Sketch", "metrics": {"acc1": 0.0024759771266874963, "acc5": 0.011574210536658218, "mean_per_class_recall": 0.0024741176470588237, "main_metric": 0.0024759771266874963}}
|
| 14 |
+
{"key": "imagenetv2", "dataset": "ImageNet v2", "metrics": {"acc1": 0.0066, "acc5": 0.0226, "mean_per_class_recall": 0.006600000000000001, "main_metric": 0.0066}}
|
| 15 |
+
{"key": "imagenet-a", "dataset": "ImageNet-A", "metrics": {"acc1": 0.008666666666666666, "acc5": 0.037866666666666667, "mean_per_class_recall": 0.009567377706780107, "main_metric": 0.008666666666666666}}
|
| 16 |
+
{"key": "imagenet-o", "dataset": "ImageNet-O", "metrics": {"acc1": 0.023, "acc5": 0.0925, "mean_per_class_recall": 0.021830666044481833, "main_metric": 0.023}}
|
| 17 |
+
{"key": "imagenet-r", "dataset": "ImageNet-R", "metrics": {"acc1": 0.014733333333333333, "acc5": 0.059533333333333334, "mean_per_class_recall": 0.013938016845053051, "main_metric": 0.014733333333333333}}
|
| 18 |
+
{"key": "vtab/kitti_closest_vehicle_distance", "dataset": "KITTI Vehicle Distance", "metrics": {"acc1": 0.22081575246132207, "acc5": null, "mean_per_class_recall": 0.25, "main_metric": 0.22081575246132207}}
|
| 19 |
+
{"key": "mnist", "dataset": "MNIST", "metrics": {"acc1": 0.0376, "acc5": 0.4196, "mean_per_class_recall": 0.03787134544976087, "main_metric": 0.0376}}
|
| 20 |
+
{"key": "objectnet", "dataset": "ObjectNet", "metrics": {"acc1": 0.013944223107569721, "acc5": 0.06363734252180467, "mean_per_class_recall": 0.014125144134702132, "main_metric": 0.013944223107569721}}
|
| 21 |
+
{"key": "vtab/flowers", "dataset": "Oxford Flowers-102", "metrics": {"acc1": 0.0065051227841925515, "acc5": 0.04456009107171898, "mean_per_class_recall": 0.006313465113208795, "main_metric": 0.006313465113208795}}
|
| 22 |
+
{"key": "vtab/pets", "dataset": "Oxford-IIIT Pet", "metrics": {"acc1": 0.03706732079585718, "acc5": 0.15290269828291086, "mean_per_class_recall": 0.0373961095085187, "main_metric": 0.0373961095085187}}
|
| 23 |
+
{"key": "voc2007", "dataset": "Pascal VOC 2007", "metrics": {"acc1": 0.08660523504273504, "acc5": 0.40197649572649574, "mean_per_class_recall": 0.1040175928424822, "main_metric": 0.08660523504273504}}
|
| 24 |
+
{"key": "vtab/pcam", "dataset": "PatchCamelyon", "metrics": {"acc1": 0.4696044921875, "acc5": null, "mean_per_class_recall": 0.4694471508372962, "main_metric": 0.4696044921875}}
|
| 25 |
+
{"key": "renderedsst2", "dataset": "Rendered SST2", "metrics": {"acc1": 0.49917627677100496, "acc5": null, "mean_per_class_recall": 0.5, "main_metric": 0.49917627677100496}}
|
| 26 |
+
{"key": "vtab/resisc45", "dataset": "RESISC45", "metrics": {"acc1": 0.032063492063492065, "acc5": 0.1376190476190476, "mean_per_class_recall": 0.03325926461796593, "main_metric": 0.032063492063492065}}
|
| 27 |
+
{"key": "cars", "dataset": "Stanford Cars", "metrics": {"acc1": 0.0034821539609501307, "acc5": 0.026737967914438502, "mean_per_class_recall": 0.0034186982184681433, "main_metric": 0.0034821539609501307}}
|
| 28 |
+
{"key": "stl10", "dataset": "STL-10", "metrics": {"acc1": 0.214875, "acc5": 0.652375, "mean_per_class_recall": 0.214875, "main_metric": 0.214875}}
|
| 29 |
+
{"key": "sun397", "dataset": "SUN397", "metrics": {"acc1": 0.013250087353108852, "acc5": 0.05388307556503669, "mean_per_class_recall": 0.010832830181802016, "main_metric": 0.013250087353108852}}
|
| 30 |
+
{"key": "vtab/svhn", "dataset": "SVHN", "metrics": {"acc1": 0.1164720344191764, "acc5": 0.49838660110633065, "mean_per_class_recall": 0.10206904240859907, "main_metric": 0.1164720344191764}}
|
| 31 |
+
{"key": "retrieval/flickr_1k_test_image_text_retrieval", "dataset": "Flickr", "metrics": {"image_retrieval_recall@1": 0.00559999980032444, "text_retrieval_recall@1": 0.006000000052154064, "image_retrieval_recall@5": 0.021400000900030136, "text_retrieval_recall@5": 0.028999999165534973, "image_retrieval_recall@10": 0.03759999945759773, "text_retrieval_recall@10": 0.04100000113248825, "mean_recall@1": 0.005799999926239252, "main_metric": 0.005799999926239252}}
|
| 32 |
+
{"key": "retrieval/mscoco_2014_5k_test_image_text_retrieval", "dataset": "MSCOCO", "metrics": {"image_retrieval_recall@1": 0.0015993602573871613, "text_retrieval_recall@1": 0.002199999988079071, "image_retrieval_recall@5": 0.007996801286935806, "text_retrieval_recall@5": 0.008799999952316284, "image_retrieval_recall@10": 0.01459416188299656, "text_retrieval_recall@10": 0.01759999990463257, "mean_recall@1": 0.0018996801227331161, "main_metric": 0.0018996801227331161}}
|
| 33 |
+
{"key": "misc/winogavil", "dataset": "WinoGAViL", "metrics": {"avg_jaccard_score": 0.2992074629458856, "jaccard_score_5": 0.3657828282828283, "jaccard_score_6": 0.3119510439164867, "jaccard_score_10": 0.18961547060138606, "jaccard_score_12": 0.1616043613707165, "jaccard_score_5-6": 0.3381813707395103, "jaccard_score_10-12": 0.17557711609233856, "main_metric": 0.17557711609233856}}
|
| 34 |
+
{"key": "wilds/iwildcam", "dataset": "iWildCam", "metrics": {"acc1": 0.0066369096305297845, "acc5": 0.0584235002687481, "mean_per_class_recall": 0.00487690046028252, "acc_avg": 0.0066369096748530865, "recall-macro_all": 0.00487690046028252, "F1-macro_all": 0.002256235295772968, "main_metric": 0.002256235295772968}}
|
| 35 |
+
{"key": "wilds/camelyon17", "dataset": "Camelyon17", "metrics": {"acc1": 0.4459402262092318, "acc5": null, "mean_per_class_recall": 0.4459402262092318, "acc_avg": 0.445940226316452, "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": 0.85196852684021, "count_slide:20": 3810.0, "acc_slide:21": 0.9106659293174744, "count_slide:21": 3694.0, "acc_slide:22": 0.32579749822616577, "count_slide:22": 7210.0, "acc_slide:23": 0.458774596452713, "count_slide:23": 5288.0, "acc_slide:24": 0.939691960811615, "count_slide:24": 7727.0, "acc_slide:25": 0.6545916199684143, "count_slide:25": 4334.0, "acc_slide:26": 0.7994757294654846, "count_slide:26": 3815.0, "acc_slide:27": 0.892010509967804, "count_slide:27": 4556.0, "acc_slide:28": 0.13015244901180267, "count_slide:28": 31878.0, "acc_slide:29": 0.40676504373550415, "count_slide:29": 12742.0, "acc_wg": 0.13015244901180267, "main_metric": 0.4459402262092318}}
|
| 36 |
+
{"key": "wilds/fmow", "dataset": "FMoW", "metrics": {"acc1": 0.02030939026596707, "acc5": 0.10150171883481093, "mean_per_class_recall": 0.01863112377694574, "acc_avg": 0.020309390500187874, "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": 0.017482297495007515, "count_year:14": 15959.0, "acc_year:15": 0.02764677256345749, "count_year:15": 6149.0, "acc_worst_year": 0.017482297495007515, "acc_region:0": 0.038081806153059006, "count_region:0": 4963.0, "acc_region:1": 0.017412086948752403, "count_region:1": 5858.0, "acc_region:2": 0.019282683730125427, "count_region:2": 2593.0, "acc_region:3": 0.012213359586894512, "count_region:3": 8024.0, "acc_region:4": 0.015015015378594398, "count_region:4": 666.0, "acc_region:5": 0.0, "count_region:5": 4.0, "acc_worst_region": 0.0, "main_metric": 0.0}}
|
| 37 |
+
{"key": "fairness/dollar_street", "dataset": "Dollar Street", "metrics": {"acc1": 0.0325435341136169, "acc5": 0.14187838995147017, "mean_per_class_recall": 0.031619802231783255, "acc_top5_avg": 0.1418783962726593, "acc_top5_income_ds:0": 0.12967289984226227, "count_income_ds:0": 856.0, "acc_top5_income_ds:1": 0.13800905644893646, "count_income_ds:1": 884.0, "acc_top5_income_ds:2": 0.14206437766551971, "count_income_ds:2": 901.0, "acc_top5_income_ds:3": 0.15777261555194855, "count_income_ds:3": 862.0, "acc_top5_wg": 0.12967289984226227, "main_metric": 0.12967289984226227}}
|
| 38 |
+
{"key": "fairness/geode", "dataset": "GeoDE", "metrics": {"acc1": 0.08920563741191544, "acc5": 0.3129404228058937, "mean_per_class_recall": 0.08945309358365802, "acc_avg": 0.08920563757419586, "acc_region:0": 0.07849686592817307, "count_region:0": 2395.0, "acc_region:1": 0.08606965094804764, "count_region:1": 2010.0, "acc_region:2": 0.08701787143945694, "count_region:2": 2126.0, "acc_region:3": 0.09347714483737946, "count_region:3": 1947.0, "acc_region:4": 0.09789413958787918, "count_region:4": 1757.0, "acc_region:5": 0.09498446434736252, "count_region:5": 2253.0, "acc_wg": 0.07849686592817307, "main_metric": 0.07849686592817307}}
|
| 39 |
+
{"key": "fairness/fairface", "dataset": "FairFace", "metrics": {"acc_race_avg": 0.4989044964313507, "acc_race_race_binary:0": 0.4513189494609833, "count_race_binary:0": 2085.0, "acc_race_race_binary:1": 0.5100913047790527, "count_race_binary:1": 8869.0, "acc_race_wg": 0.4513189494609833, "acc_gender_avg": 0.5149716734886169, "acc_gender_race_binary:0": 0.5247002243995667, "acc_gender_race_binary:1": 0.5126846432685852, "acc_gender_wg": 0.5126846432685852, "acc_age_avg": 0.024009494110941887, "acc_age_race_binary:0": 0.026378896087408066, "acc_age_race_binary:1": 0.023452475666999817, "acc_age_wg": 0.023452475666999817, "acc_gender_x_avg": 0.5149716734886169, "acc_gender_x_race:0_gender:0": 0.15018773078918457, "count_race:0_gender:0": 799.0, "acc_gender_x_race:0_gender:1": 0.8494055271148682, "count_race:0_gender:1": 757.0, "acc_gender_x_race:1_gender:0": 0.2869875133037567, "count_race:1_gender:0": 1122.0, "acc_gender_x_race:1_gender:1": 0.801661491394043, "count_race:1_gender:1": 963.0, "acc_gender_x_race:2_gender:0": 0.17795485258102417, "count_race:2_gender:0": 753.0, "acc_gender_x_race:2_gender:1": 0.8414154648780823, "count_race:2_gender:1": 763.0, "acc_gender_x_race:3_gender:0": 0.2786885201931, "count_race:3_gender:0": 793.0, "acc_gender_x_race:3_gender:1": 0.8072289228439331, "count_race:3_gender:1": 830.0, "acc_gender_x_race:4_gender:0": 0.2632226347923279, "count_race:4_gender:0": 813.0, "acc_gender_x_race:4_gender:1": 0.7803030014038086, "count_race:4_gender:1": 396.0, "acc_gender_x_race:5_gender:0": 0.25170066952705383, "count_race:5_gender:0": 735.0, "acc_gender_x_race:5_gender:1": 0.7926470637321472, "count_race:5_gender:1": 680.0, "acc_gender_x_race:6_gender:0": 0.28828829526901245, "count_race:6_gender:0": 777.0, "acc_gender_x_race:6_gender:1": 0.835705041885376, "count_race:6_gender:1": 773.0, "acc_gender_x_wg": 0.15018773078918457, "toxicity_crime_avg": 0.034781813621520996, "toxicity_crime_race:0": 0.057197943329811096, "count_race:0": 1556.0, "toxicity_crime_race:1": 0.036450840532779694, "count_race:1": 2085.0, "toxicity_crime_race:2": 0.036939315497875214, "count_race:2": 1516.0, "toxicity_crime_race:3": 0.019716573879122734, "count_race:3": 1623.0, "toxicity_crime_race:4": 0.038047973066568375, "count_race:4": 1209.0, "toxicity_crime_race:5": 0.037455830723047256, "count_race:5": 1415.0, "toxicity_crime_race:6": 0.018709678202867508, "count_race:6": 1550.0, "toxicity_crime_wg": 0.018709678202867508, "toxicity_nonhuman_avg": 0.2090560495853424, "toxicity_nonhuman_race:0": 0.15038560330867767, "toxicity_nonhuman_race:1": 0.23117506504058838, "toxicity_nonhuman_race:2": 0.17810025811195374, "toxicity_nonhuman_race:3": 0.23105360567569733, "toxicity_nonhuman_race:4": 0.20430107414722443, "toxicity_nonhuman_race:5": 0.2233215570449829, "toxicity_nonhuman_race:6": 0.236129030585289, "toxicity_nonhuman_wg": 0.15038560330867767, "main_metric": null}}
|
| 40 |
+
{"key": "fairness/utkface", "dataset": "UTKFace", "metrics": {"acc_race_avg": 0.5062650442123413, "acc_race_race_binary:0": 0.5629218220710754, "count_race_binary:0": 10076.0, "acc_race_race_binary:1": 0.4643721878528595, "count_race_binary:1": 13627.0, "acc_race_wg": 0.4643721878528595, "acc_gender_avg": 0.5225498676300049, "acc_gender_race_binary:0": 0.5242159366607666, "acc_gender_race_binary:1": 0.5213179588317871, "acc_gender_wg": 0.5213179588317871, "acc_age_avg": 0.06134244427084923, "acc_age_race_binary:0": 0.042080190032720566, "acc_age_race_binary:1": 0.07558523863554001, "acc_age_wg": 0.042080190032720566, "acc_gender_x_avg": 0.5225498676300049, "acc_gender_x_race:0_gender:0": 0.14106988906860352, "count_race:0_gender:0": 2318.0, "acc_gender_x_race:0_gender:1": 0.8405796885490417, "count_race:0_gender:1": 2208.0, "acc_gender_x_race:1_gender:0": 0.2205989807844162, "count_race:1_gender:0": 5476.0, "acc_gender_x_race:1_gender:1": 0.8856521844863892, "count_race:1_gender:1": 4600.0, "acc_gender_x_race:2_gender:0": 0.23308271169662476, "count_race:2_gender:0": 2261.0, "acc_gender_x_race:2_gender:1": 0.8541423678398132, "count_race:2_gender:1": 1714.0, "acc_gender_x_race:3_gender:0": 0.19174602627754211, "count_race:3_gender:0": 1575.0, "acc_gender_x_race:3_gender:1": 0.8795050978660583, "count_race:3_gender:1": 1859.0, "acc_gender_x_race:4_gender:0": 0.19736842811107635, "count_race:4_gender:0": 760.0, "acc_gender_x_race:4_gender:1": 0.9045064449310303, "count_race:4_gender:1": 932.0, "acc_gender_x_wg": 0.14106988906860352, "toxicity_crime_avg": 0.009492469020187855, "toxicity_crime_race:0": 0.020768890157341957, "count_race:0": 4526.0, "toxicity_crime_race:1": 0.007741167210042477, "count_race:1": 10076.0, "toxicity_crime_race:2": 0.0052830190397799015, "count_race:2": 3975.0, "toxicity_crime_race:3": 0.005532906390726566, "count_race:3": 3434.0, "toxicity_crime_race:4": 0.007683215197175741, "count_race:4": 1692.0, "toxicity_crime_wg": 0.0052830190397799015, "toxicity_nonhuman_avg": 0.3261612355709076, "toxicity_nonhuman_race:0": 0.29032257199287415, "toxicity_nonhuman_race:1": 0.34051209688186646, "toxicity_nonhuman_race:2": 0.3708176016807556, "toxicity_nonhuman_race:3": 0.27402445673942566, "toxicity_nonhuman_race:4": 0.3374704420566559, "toxicity_nonhuman_wg": 0.27402445673942566, "main_metric": null}}
|
minipile_style_trained/minipile_style/eval_results.jsonl
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{"key": "vtab/caltech101", "dataset": "Caltech-101", "metrics": {"acc1": 0.05801150369761709, "acc5": 0.2325390304026294, "mean_per_class_recall": 0.09916717589448028, "main_metric": 0.09916717589448028}}
|
| 2 |
+
{"key": "cifar10", "dataset": "CIFAR-10", "metrics": {"acc1": 0.3137, "acc5": 0.8126, "mean_per_class_recall": 0.3137, "main_metric": 0.3137}}
|
| 3 |
+
{"key": "vtab/cifar100", "dataset": "CIFAR-100", "metrics": {"acc1": 0.1024, "acc5": 0.2848, "mean_per_class_recall": 0.1024, "main_metric": 0.1024}}
|
| 4 |
+
{"key": "vtab/clevr_count_all", "dataset": "CLEVR Counts", "metrics": {"acc1": 0.1256, "acc5": 0.5966666666666667, "mean_per_class_recall": 0.12539439963637974, "main_metric": 0.1256}}
|
| 5 |
+
{"key": "vtab/clevr_closest_object_distance", "dataset": "CLEVR Distance", "metrics": {"acc1": 0.19546666666666668, "acc5": 0.9186666666666666, "mean_per_class_recall": 0.15450034668863621, "main_metric": 0.19546666666666668}}
|
| 6 |
+
{"key": "country211", "dataset": "Country211", "metrics": {"acc1": 0.01052132701421801, "acc5": 0.04303317535545024, "mean_per_class_recall": 0.010521327014218011, "main_metric": 0.01052132701421801}}
|
| 7 |
+
{"key": "vtab/dtd", "dataset": "Describable Textures", "metrics": {"acc1": 0.03031914893617021, "acc5": 0.12659574468085105, "mean_per_class_recall": 0.030319148936170215, "main_metric": 0.03031914893617021}}
|
| 8 |
+
{"key": "vtab/eurosat", "dataset": "EuroSAT", "metrics": {"acc1": 0.11222222222222222, "acc5": 0.5272222222222223, "mean_per_class_recall": 0.11740399043932866, "main_metric": 0.11222222222222222}}
|
| 9 |
+
{"key": "fgvc_aircraft", "dataset": "FGVC Aircraft", "metrics": {"acc1": 0.0111011101110111, "acc5": 0.058205820582058206, "mean_per_class_recall": 0.010962566844919787, "main_metric": 0.010962566844919787}}
|
| 10 |
+
{"key": "food101", "dataset": "Food-101", "metrics": {"acc1": 0.048514851485148516, "acc5": 0.16784158415841585, "mean_per_class_recall": 0.048514851485148516, "main_metric": 0.048514851485148516}}
|
| 11 |
+
{"key": "gtsrb", "dataset": "GTSRB", "metrics": {"acc1": 0.034996041171813146, "acc5": 0.15273159144893111, "mean_per_class_recall": 0.0325462314807613, "main_metric": 0.034996041171813146}}
|
| 12 |
+
{"key": "imagenet1k", "dataset": "ImageNet 1k", "metrics": {"acc1": 0.02072, "acc5": 0.06912, "mean_per_class_recall": 0.02072, "main_metric": 0.02072}}
|
| 13 |
+
{"key": "imagenet_sketch", "dataset": "ImageNet Sketch", "metrics": {"acc1": 0.005639725677454853, "acc5": 0.022008685570555523, "mean_per_class_recall": 0.005712941176470588, "main_metric": 0.005639725677454853}}
|
| 14 |
+
{"key": "imagenetv2", "dataset": "ImageNet v2", "metrics": {"acc1": 0.0184, "acc5": 0.0577, "mean_per_class_recall": 0.018400000000000003, "main_metric": 0.0184}}
|
| 15 |
+
{"key": "imagenet-a", "dataset": "ImageNet-A", "metrics": {"acc1": 0.011066666666666667, "acc5": 0.05333333333333334, "mean_per_class_recall": 0.013485098560340742, "main_metric": 0.011066666666666667}}
|
| 16 |
+
{"key": "imagenet-o", "dataset": "ImageNet-O", "metrics": {"acc1": 0.068, "acc5": 0.192, "mean_per_class_recall": 0.06090153207645468, "main_metric": 0.068}}
|
| 17 |
+
{"key": "imagenet-r", "dataset": "ImageNet-R", "metrics": {"acc1": 0.0242, "acc5": 0.08956666666666667, "mean_per_class_recall": 0.02462999037160764, "main_metric": 0.0242}}
|
| 18 |
+
{"key": "vtab/kitti_closest_vehicle_distance", "dataset": "KITTI Vehicle Distance", "metrics": {"acc1": 0.18706047819971872, "acc5": null, "mean_per_class_recall": 0.21116877016445879, "main_metric": 0.18706047819971872}}
|
| 19 |
+
{"key": "mnist", "dataset": "MNIST", "metrics": {"acc1": 0.1043, "acc5": 0.4795, "mean_per_class_recall": 0.10318325491005997, "main_metric": 0.1043}}
|
| 20 |
+
{"key": "objectnet", "dataset": "ObjectNet", "metrics": {"acc1": 0.022558414988693872, "acc5": 0.08937224076666307, "mean_per_class_recall": 0.024320416146038346, "main_metric": 0.022558414988693872}}
|
| 21 |
+
{"key": "vtab/flowers", "dataset": "Oxford Flowers-102", "metrics": {"acc1": 0.02569523499756058, "acc5": 0.09708895755407383, "mean_per_class_recall": 0.025834503214636727, "main_metric": 0.025834503214636727}}
|
| 22 |
+
{"key": "vtab/pets", "dataset": "Oxford-IIIT Pet", "metrics": {"acc1": 0.039247751430907606, "acc5": 0.1373671300081766, "mean_per_class_recall": 0.03911935815147742, "main_metric": 0.03911935815147742}}
|
| 23 |
+
{"key": "voc2007", "dataset": "Pascal VOC 2007", "metrics": {"acc1": 0.14349626068376067, "acc5": 0.49993322649572647, "mean_per_class_recall": 0.165779401548417, "main_metric": 0.14349626068376067}}
|
| 24 |
+
{"key": "vtab/pcam", "dataset": "PatchCamelyon", "metrics": {"acc1": 0.494598388671875, "acc5": null, "mean_per_class_recall": 0.49439772488731343, "main_metric": 0.494598388671875}}
|
| 25 |
+
{"key": "renderedsst2", "dataset": "Rendered SST2", "metrics": {"acc1": 0.4986271279516749, "acc5": null, "mean_per_class_recall": 0.49806033234902436, "main_metric": 0.4986271279516749}}
|
| 26 |
+
{"key": "vtab/resisc45", "dataset": "RESISC45", "metrics": {"acc1": 0.039047619047619046, "acc5": 0.17555555555555555, "mean_per_class_recall": 0.03993347555543864, "main_metric": 0.039047619047619046}}
|
| 27 |
+
{"key": "cars", "dataset": "Stanford Cars", "metrics": {"acc1": 0.012560626787712971, "acc5": 0.050242507150851885, "mean_per_class_recall": 0.012540794922037909, "main_metric": 0.012560626787712971}}
|
| 28 |
+
{"key": "stl10", "dataset": "STL-10", "metrics": {"acc1": 0.3605, "acc5": 0.86125, "mean_per_class_recall": 0.3605, "main_metric": 0.3605}}
|
| 29 |
+
{"key": "sun397", "dataset": "SUN397", "metrics": {"acc1": 0.04235246519668242, "acc5": 0.12996303584235983, "mean_per_class_recall": 0.034176297735869304, "main_metric": 0.04235246519668242}}
|
| 30 |
+
{"key": "vtab/svhn", "dataset": "SVHN", "metrics": {"acc1": 0.06906883835279656, "acc5": 0.40104486785494775, "mean_per_class_recall": 0.08868828572041501, "main_metric": 0.06906883835279656}}
|
| 31 |
+
{"key": "retrieval/flickr_1k_test_image_text_retrieval", "dataset": "Flickr", "metrics": {"image_retrieval_recall@1": 0.009999999776482582, "text_retrieval_recall@1": 0.012000000104308128, "image_retrieval_recall@5": 0.03240000084042549, "text_retrieval_recall@5": 0.03999999910593033, "image_retrieval_recall@10": 0.0575999990105629, "text_retrieval_recall@10": 0.05900000035762787, "mean_recall@1": 0.010999999940395355, "main_metric": 0.010999999940395355}}
|
| 32 |
+
{"key": "retrieval/mscoco_2014_5k_test_image_text_retrieval", "dataset": "MSCOCO", "metrics": {"image_retrieval_recall@1": 0.003998400643467903, "text_retrieval_recall@1": 0.005200000014156103, "image_retrieval_recall@5": 0.016393441706895828, "text_retrieval_recall@5": 0.01860000006854534, "image_retrieval_recall@10": 0.03126749396324158, "text_retrieval_recall@10": 0.03099999949336052, "mean_recall@1": 0.004599200328812003, "main_metric": 0.004599200328812003}}
|
| 33 |
+
{"key": "misc/winogavil", "dataset": "WinoGAViL", "metrics": {"avg_jaccard_score": 0.32688745439236594, "jaccard_score_5": 0.3870959595959596, "jaccard_score_6": 0.3381089512838973, "jaccard_score_10": 0.22506147999105744, "jaccard_score_12": 0.2061303960836671, "jaccard_score_5-6": 0.36197858988556664, "jaccard_score_10-12": 0.21557377049180326, "main_metric": 0.21557377049180326}}
|
| 34 |
+
{"key": "wilds/iwildcam", "dataset": "iWildCam", "metrics": {"acc1": 0.004089645018812367, "acc5": 0.036175831366408825, "mean_per_class_recall": 0.008390848546851286, "acc_avg": 0.004089645110070705, "recall-macro_all": 0.008390848546851286, "F1-macro_all": 0.0036140537071774766, "main_metric": 0.0036140537071774766}}
|
| 35 |
+
{"key": "wilds/camelyon17", "dataset": "Camelyon17", "metrics": {"acc1": 0.49563806522914855, "acc5": null, "mean_per_class_recall": 0.49563806522914855, "acc_avg": 0.49563807249069214, "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": 0.9808399081230164, "count_slide:20": 3810.0, "acc_slide:21": 0.9940444231033325, "count_slide:21": 3694.0, "acc_slide:22": 0.4098474383354187, "count_slide:22": 7210.0, "acc_slide:23": 0.47617247700691223, "count_slide:23": 5288.0, "acc_slide:24": 0.9708813428878784, "count_slide:24": 7727.0, "acc_slide:25": 0.7512690424919128, "count_slide:25": 4334.0, "acc_slide:26": 0.861074686050415, "count_slide:26": 3815.0, "acc_slide:27": 0.9712467193603516, "count_slide:27": 4556.0, "acc_slide:28": 0.15173473954200745, "count_slide:28": 31878.0, "acc_slide:29": 0.46845078468322754, "count_slide:29": 12742.0, "acc_wg": 0.15173473954200745, "main_metric": 0.49563806522914855}}
|
| 36 |
+
{"key": "wilds/fmow", "dataset": "FMoW", "metrics": {"acc1": 0.016012303238646643, "acc5": 0.07770942645196309, "mean_per_class_recall": 0.01662464627996616, "acc_avg": 0.01601230353116989, "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": 0.015289178118109703, "count_year:14": 15959.0, "acc_year:15": 0.01788908801972866, "count_year:15": 6149.0, "acc_worst_year": 0.015289178118109703, "acc_region:0": 0.025387870147824287, "count_region:0": 4963.0, "acc_region:1": 0.014339365065097809, "count_region:1": 5858.0, "acc_region:2": 0.008098727092146873, "count_region:2": 2593.0, "acc_region:3": 0.01296111661940813, "count_region:3": 8024.0, "acc_region:4": 0.028528528288006783, "count_region:4": 666.0, "acc_region:5": 0.0, "count_region:5": 4.0, "acc_worst_region": 0.0, "main_metric": 0.0}}
|
| 37 |
+
{"key": "fairness/dollar_street", "dataset": "Dollar Street", "metrics": {"acc1": 0.06365972023979447, "acc5": 0.22723379960034257, "mean_per_class_recall": 0.06571056459525114, "acc_top5_avg": 0.22723379731178284, "acc_top5_income_ds:0": 0.18691588938236237, "count_income_ds:0": 856.0, "acc_top5_income_ds:1": 0.20135746896266937, "count_income_ds:1": 884.0, "acc_top5_income_ds:2": 0.23640400171279907, "count_income_ds:2": 901.0, "acc_top5_income_ds:3": 0.2842227518558502, "count_income_ds:3": 862.0, "acc_top5_wg": 0.18691588938236237, "main_metric": 0.18691588938236237}}
|
| 38 |
+
{"key": "fairness/geode", "dataset": "GeoDE", "metrics": {"acc1": 0.15783151825752723, "acc5": 0.4217648942985266, "mean_per_class_recall": 0.1539005317667282, "acc_avg": 0.15783151984214783, "acc_region:0": 0.15198330581188202, "count_region:0": 2395.0, "acc_region:1": 0.14676617085933685, "count_region:1": 2010.0, "acc_region:2": 0.15522107481956482, "count_region:2": 2126.0, "acc_region:3": 0.1587056964635849, "count_region:3": 1947.0, "acc_region:4": 0.15139442682266235, "count_region:4": 1757.0, "acc_region:5": 0.18064802885055542, "count_region:5": 2253.0, "acc_wg": 0.14676617085933685, "main_metric": 0.14676617085933685}}
|
| 39 |
+
{"key": "fairness/fairface", "dataset": "FairFace", "metrics": {"acc_race_avg": 0.6924411058425903, "acc_race_race_binary:0": 0.18896882236003876, "count_race_binary:0": 2085.0, "acc_race_race_binary:1": 0.8108016848564148, "count_race_binary:1": 8869.0, "acc_race_wg": 0.18896882236003876, "acc_gender_avg": 0.4889537990093231, "acc_gender_race_binary:0": 0.4657074213027954, "acc_gender_race_binary:1": 0.49441877007484436, "acc_gender_wg": 0.4657074213027954, "acc_age_avg": 0.053313858807086945, "acc_age_race_binary:0": 0.042206235229969025, "acc_age_race_binary:1": 0.05592513084411621, "acc_age_wg": 0.042206235229969025, "acc_gender_x_avg": 0.4889537990093231, "acc_gender_x_race:0_gender:0": 0.2152690887451172, "count_race:0_gender:0": 799.0, "acc_gender_x_race:0_gender:1": 0.7859973311424255, "count_race:0_gender:1": 757.0, "acc_gender_x_race:1_gender:0": 0.28431373834609985, "count_race:1_gender:0": 1122.0, "acc_gender_x_race:1_gender:1": 0.6770508885383606, "count_race:1_gender:1": 963.0, "acc_gender_x_race:2_gender:0": 0.22045153379440308, "count_race:2_gender:0": 753.0, "acc_gender_x_race:2_gender:1": 0.7876802086830139, "count_race:2_gender:1": 763.0, "acc_gender_x_race:3_gender:0": 0.3366960883140564, "count_race:3_gender:0": 793.0, "acc_gender_x_race:3_gender:1": 0.6457831263542175, "count_race:3_gender:1": 830.0, "acc_gender_x_race:4_gender:0": 0.2964329719543457, "count_race:4_gender:0": 813.0, "acc_gender_x_race:4_gender:1": 0.6792929172515869, "count_race:4_gender:1": 396.0, "acc_gender_x_race:5_gender:0": 0.31700679659843445, "count_race:5_gender:0": 735.0, "acc_gender_x_race:5_gender:1": 0.7029411792755127, "count_race:5_gender:1": 680.0, "acc_gender_x_race:6_gender:0": 0.35521236062049866, "count_race:6_gender:0": 777.0, "acc_gender_x_race:6_gender:1": 0.7128072381019592, "count_race:6_gender:1": 773.0, "acc_gender_x_wg": 0.2152690887451172, "toxicity_crime_avg": 0.12853752076625824, "toxicity_crime_race:0": 0.09640102833509445, "count_race:0": 1556.0, "toxicity_crime_race:1": 0.14196643233299255, "count_race:1": 2085.0, "toxicity_crime_race:2": 0.10554089397192001, "count_race:2": 1516.0, "toxicity_crime_race:3": 0.14356130361557007, "count_race:3": 1623.0, "toxicity_crime_race:4": 0.11166252940893173, "count_race:4": 1209.0, "toxicity_crime_race:5": 0.13144876062870026, "count_race:5": 1415.0, "toxicity_crime_race:6": 0.1599999964237213, "count_race:6": 1550.0, "toxicity_crime_wg": 0.09640102833509445, "toxicity_nonhuman_avg": 0.3316596746444702, "toxicity_nonhuman_race:0": 0.3971722424030304, "toxicity_nonhuman_race:1": 0.3122302293777466, "toxicity_nonhuman_race:2": 0.35224273800849915, "toxicity_nonhuman_race:3": 0.3425754904747009, "toxicity_nonhuman_race:4": 0.3325062096118927, "toxicity_nonhuman_race:5": 0.30459365248680115, "toxicity_nonhuman_race:6": 0.2845161259174347, "toxicity_nonhuman_wg": 0.2845161259174347, "main_metric": null}}
|
| 40 |
+
{"key": "fairness/utkface", "dataset": "UTKFace", "metrics": {"acc_race_avg": 0.5589587688446045, "acc_race_race_binary:0": 0.15452560782432556, "count_race_binary:0": 10076.0, "acc_race_race_binary:1": 0.8580024838447571, "count_race_binary:1": 13627.0, "acc_race_wg": 0.15452560782432556, "acc_gender_avg": 0.49516940116882324, "acc_gender_race_binary:0": 0.46724891662597656, "acc_gender_race_binary:1": 0.5158141851425171, "acc_gender_wg": 0.46724891662597656, "acc_age_avg": 0.0861494317650795, "acc_age_race_binary:0": 0.11601825803518295, "acc_age_race_binary:1": 0.06406398862600327, "acc_age_wg": 0.06406398862600327, "acc_gender_x_avg": 0.49516940116882324, "acc_gender_x_race:0_gender:0": 0.44305434823036194, "count_race:0_gender:0": 2318.0, "acc_gender_x_race:0_gender:1": 0.6014492511749268, "count_race:0_gender:1": 2208.0, "acc_gender_x_race:1_gender:0": 0.5776113867759705, "count_race:1_gender:0": 5476.0, "acc_gender_x_race:1_gender:1": 0.33586955070495605, "count_race:1_gender:1": 4600.0, "acc_gender_x_race:2_gender:0": 0.6112339496612549, "count_race:2_gender:0": 2261.0, "acc_gender_x_race:2_gender:1": 0.39789965748786926, "count_race:2_gender:1": 1714.0, "acc_gender_x_race:3_gender:0": 0.6793650984764099, "count_race:3_gender:0": 1575.0, "acc_gender_x_race:3_gender:1": 0.37116730213165283, "count_race:3_gender:1": 1859.0, "acc_gender_x_race:4_gender:0": 0.6723684072494507, "count_race:4_gender:0": 760.0, "acc_gender_x_race:4_gender:1": 0.3637339174747467, "count_race:4_gender:1": 932.0, "acc_gender_x_wg": 0.33586955070495605, "toxicity_crime_avg": 0.16407205164432526, "toxicity_crime_race:0": 0.13676536083221436, "count_race:0": 4526.0, "toxicity_crime_race:1": 0.19213974475860596, "count_race:1": 10076.0, "toxicity_crime_race:2": 0.13735848665237427, "count_race:2": 3975.0, "toxicity_crime_race:3": 0.14967967569828033, "count_race:3": 3434.0, "toxicity_crime_race:4": 0.16193853318691254, "count_race:4": 1692.0, "toxicity_crime_wg": 0.13676536083221436, "toxicity_nonhuman_avg": 0.4111293852329254, "toxicity_nonhuman_race:0": 0.5141405463218689, "toxicity_nonhuman_race:1": 0.3814013600349426, "toxicity_nonhuman_race:2": 0.4646540880203247, "toxicity_nonhuman_race:3": 0.3046010434627533, "toxicity_nonhuman_race:4": 0.4030732810497284, "toxicity_nonhuman_wg": 0.3046010434627533, "main_metric": null}}
|