search

zhanghang1989 / ResNeSt

ResNeSt: Split-Attention Networks
https://arxiv.org/abs/2004.08955
Apache License 2.0
3.22k stars 496 forks source link

deeplabv3 #136

Open Cyan0FF opened 3 years ago

Cyan0FF commented 3 years ago

Hello, I use resnest50-deeplabv3 model and train from scratch on Cityscapes dataset. During training, the loss value doesn't decrease after reaching 0.1( I use a single titan gpu and batch size of 10 ). Is this normal ? and can i get an ok result using single gpu ？

zhanghang1989 commented 3 years ago

Cityscapes requires larger crop size, which is very memory consuming. Unfortunately, single GPU won't work