I am reproducing your results using the pretrained ResNet50 you provided on the RSNA dataset. When training with 1% training data, I only got 59%, which is lower than the 63% Dice in your paper on the RSNA dataset.
The settings are as follows:
learning rate: 5e-4, batch size: 16, img_size: 512, weight_decay: 1e-6, seed: 42, num_workers: 16.
I ran several times but the results remain the same. Do you have any suggestions for this problem?
Hello @fuying-wang ,
Thank you for your code.
I am reproducing your results using the pretrained ResNet50 you provided on the RSNA dataset. When training with 1% training data, I only got 59%, which is lower than the 63% Dice in your paper on the RSNA dataset. The settings are as follows: learning rate: 5e-4, batch size: 16, img_size: 512, weight_decay: 1e-6, seed: 42, num_workers: 16.
I ran several times but the results remain the same. Do you have any suggestions for this problem?
Thank you very much.