search

xuebinqin / U-2-Net

The code for our newly accepted paper in Pattern Recognition 2020: "U^2-Net: Going Deeper with Nested U-Structure for Salient Object Detection."
Apache License 2.0
8.49k stars 1.46k forks source link

About setting 'num_epoch' #237

Open dzhyy opened 3 years ago

dzhyy commented 3 years ago

Hello, I want to use the pre training model for fine-tune to adapt to my portrait matting task. I found the original value of epoch is set to a very large value (100000). If it is fine-tune on the original basis, how much 'num_epoch' set is appropriate? thank you for answer.

baleksey commented 3 years ago

@dzhyy I've trained my last model for portraits for about 130-150 epoches until it just stops decreasing loss. 100k is definitely too much. Meaningful results are already after 50-70 epoches.

The final training loss was ~ 0.146 and portrait segmentation is pretty good. So you can use those numbers as reference for your training.