downstream task of ACL using the source code and data of this repo.

[jurader]

Thank you for the source code and data of downstream tasks.

I tried to train the following command for the ACL task.

python acl_train.py --gpu_node 0 --database ImageNet --model_name ResNet50 --structure unfreezeall --batch_size 64 --lr 0.0001

However, the val AUC was lower than 0.85 for the fold=1.
I found the strong overfitting.

Please show me the command for reproducing the reuslts of your paper, for the ACL task.