search

zhanghang1989 / ResNeSt

ResNeSt: Split-Attention Networks
https://arxiv.org/abs/2004.08955
Apache License 2.0
3.23k stars 496 forks source link

backbone #99

Open lijain opened 4 years ago

lijain commented 4 years ago

Hello, I will replace the backbone network of RESNET with your resnest. During the training process, the loss value decreases very slowly (LR here is the same as that of RESNET). Similarly, in 20 epoches, the loss of RESNET is 0.06 and that of resnest is about 0.5. What's going on next?

zhanghang1989 commented 4 years ago

ResNeSt need to use large batch size per gpu or SyncBatchNorm.

Dpw506 commented 3 years ago

Hi, What is the process of using RESNETST instead of RENET?

zhanghang1989 commented 3 years ago

Hi, What is the process of using RESNETST instead of RENET?

https://github.com/zhanghang1989/ResNeSt#pytorch-models