search

BunnySoCrazy / SECAD-Net

This repository provides the official code of SECAD-Net.
MIT License
35 stars 6 forks source link

Not converging? #1

Closed kimren227 closed 4 months ago

kimren227 commented 1 year ago

Hi,

Thanks for this exciting work and for releasing the code. I tried to train the network following the instructions in the README, however, it seems that the network is not converging (see Figs attached). Any suggestions on how to train the network correctly?

Thanks,

Daxuan

image image image image image
BunnySoCrazy commented 1 year ago

I'll recheck the training code. Before that, please use the pre-training parameters in this link to fine-tune the shape. Hope this helps.

BunnySoCrazy commented 1 year ago

I discovered that the original training code was loading the test dataset instead of the training dataset. I have rectified the issue. I sincerely appreciate your valuable feedback on my mistake.

kimren227 commented 1 year ago

Thanks for the swift reply, the provided pre-trained model works well for fine-tuning. However, I still facing the same issue during training after pulling the latest change. Looking forward to the code fix. If there is anything I can help with, do let me know.

image
BunnySoCrazy commented 4 months ago

I've rerun the training code and everything seems to be working fine, so I'm going to close this issue now.