Cannot Reproduce the reported performance

[leesb7426]

I trained '1 Pre-train the First Transformer' following your instruction, but it could not be reproduced.

I utilized the below script to train the fist transformer and 480x480 resized images. 'python3 -m torch.distributed.launch --nproc_per_node=4 main_hct_first.py --arch vit_small --data_path /project/ssd2/subeen/FSC_datasets/mini-ImageNet_HCTransformers/train --output_dir save_dir_HCTransformers'

I have attached the train.log and evaluation results. Simply, it is as follows: {"train_loss": 2.3388298997779686, "train_dinoloss": 2.328510658343633, "train_clsloss": 0.010319240135334742, "train_lr": 1.012717730523827e-06, "train_wd": 0.3999981448259593, "epoch": 399}

2024-04-18 04:48:41 393 Test Acc at 10000= 62.98% +- 0.16% student

test_log_mini_student.txt log.txt args.txt