finetune with worse result

[DENGBOYU-REX]

Hi,

Thank you for your great work. I noticed that you released different .pth files. Are these from different tasks?

Also, I fine-tuned viteraser-base-withsegmim on my personal dataset but obtained worse results compared to the pretrained viteraser-base.pth. I noticed that your script loads both the encoder and the pretrained model, instead of just viteraser-base.pth. Could you please explain the reasoning behind this?

Thank you!