Unexpected result when trained resnet50 on MS1MV3 dataset

[jinkun-hao]

Hi， I trained resnet50 on MS1MV3 dataset and used the same params as given, but got 0.8882 test_acc on test dataset.

Detailed result: {'agedb_30_num_test_samples': 12000.0, 'agedb_30_test_acc': 0.8686666488647461, 'agedb_30_test_best_threshold': 1.7070000171661377, 'cfp_fp_num_test_samples': 14000.0, 'cfp_fp_test_acc': 0.8317142724990845, 'cfp_fp_test_best_threshold': 1.7899999618530273, 'lfw_num_test_samples': 12000.0, 'lfw_test_acc': 0.9645000100135803, 'lfw_test_best_threshold': 1.5600000619888306, 'test_acc': 0.8882936239242554}

params: --arch ir_50 \ --use_16bit \ --batch_size 256 \ --num_workers 8 \ --epochs 50 \ --lr_milestones 12,20,24 \ --lr 0.1 \ --head adaface \ --m 0.4 \ --h 0.333 \ --low_res_augmentation_prob 0.2 \ --crop_augmentation_prob 0.2 \ --photometric_augmentation_prob 0.2

[mk-minchul]

Hi jinkun-hao, When I tested the above configuration by training for few epochs, I get up the val_acc to be 0.9644. I think there could be a discrepancy either in the env setting or the dataset.