Detail:Qiita written in Japanese
-
Clone it.
-
You have to download CamVid dataset from https://github.com/alexgkendall/SegNet-Tutorial. Make "dataset" directory and put them into the directory.
-
You can change implementation details in config.py.
-
Run by
python3 train.py
and "saved_models" directory will be made.
-
Test by
python3 test.py
and "predicted_imgs" directory will be made. You can see predicted semantic segmantation images in the directory.
If you want to use U-Net, you have to change
network.SegNet()
to
network.UNet()
in train.py and test.py.