The problem for training

WXinlong / SOLO

SOLO and SOLOv2 for instance segmentation, ECCV 2020 & NeurIPS 2020.

Other

1.69k stars 307 forks source link

The problem for training #113

Open OpencvW opened 3 years ago

OpencvW commented 3 years ago

Hi，Thanks for your great work. I want to show the mAP value in training process， could you tell me how to change the code to show mAP also? Thank you very much.

OpencvW commented 3 years ago

I try to use the --validate parameter，does it takes a long time for validation? Why it stop there?

WXinlong commented 3 years ago

@OpencvW The --validate flag is not supported yet. You need to modify some code if you want to use it. It is supported in my implementations on mmdetv2: https://github.com/WXinlong/mmdetection/tree/SOLO.

OpencvW commented 3 years ago

Thanks for your reply. It shown this problem when I used dcnv2. Could you give me some advice？

WXinlong commented 3 years ago

@OpencvW It's OK. You can continue training.

OpencvW commented 3 years ago

Thank you very much!

OpencvW commented 3 years ago

Hi，when I added dcnv2 to the model structure, the instance loss decreased at first, but then increased. The performance of the model was worse than that without dcnv2. Is there something wrong with my settings?

HuygheB commented 2 years ago

@WXinlong is there some guidance as to what code has to be changed to enable the --validate flag?