Finetuning on my data

[DDR-noise]

I’m trying to finetune FRCRN on my own data, but the results end up worse than on original checkpoint. How can i understand learning is start from checkpoint? I set path to your model in config file. Maybe I’m doing something wrong?