mmSegmentation的单GPU训练和测试

mmSegmentation的使用

参考：

1. github

mmseg安装：

1. 安装依赖包：

python==3.6, 3.7
1.1.2<=mmcv<=1.2.0
torch==1.5.0 
torchvision==0.6.0

2. mmcv 安装

CUDA	torch 1.6	torch 1.5	torch 1.4	torch 1.3
10.2	O	O	X	X
10.1	O	O	O	O
9.2	O	O	O	O
cpu	O	O	O	O

pip install mmcv-full==latest+torch1.6.0+cu102 -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html
pip install mmcv-full==latest+torch1.5.0+cu102 -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html
pip install mmcv-full==latest+torch1.6.0+cu101 -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html
pip install mmcv-full==latest+torch1.5.0+cu101 -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html
pip install mmcv-full==latest+torch1.4.0+cu101 -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html
pip install mmcv-full==latest+torch1.3.0+cu101 -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html
pip install mmcv-full==latest+torch1.6.0+cu92 -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html
pip install mmcv-full==latest+torch1.6.0+cpu -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html

# 安装mmcv-full build时间较长
pip isntall -r requirements.txt
MMCV_WITH_OPS=1 pip install -e .

3.mmseg安装

pip install -r requirements/build.txt
pip install -e .

mmseg 训练

1. 单GPU

需要注意问题：

https://github.com/open-mmlab/mmdetection/issues/847.  SyncBN only works with distributed environment, you may either use tools/dist_test.sh or modify SyncBN to BN manually during testing.

修改方法：
将configs/_base_中的参数设置 SyncBN修改为BN

2. 多GPU

参考：github