Regarding test datset #2
Description
I used your pretrained weights (deraining.pth) to test the Restormer base model on test.py file and achieved a PSNR of 22. However, when I trained the model with pretrained weights with the training code and used the checkpoints in the testing code to resume from an epoch, I achieved a PSNR of 23. Why is the PSNR lower compared to your reported PSNR of 25?