PyTorch-Encoding

created by Hang Zhang

Documentation

• Please visit the Docs for detail instructions of installation and usage.

• Please visit the link to examples of semantic segmentation.

Install Package

1. This package relies on PyTorch master branch (higher than stable released v0.4.0), please follow the instruction to install PyTorch from source. This PyTorch master branch need cuda9.0 or higher cuda version. Otherwise, the encoding.py and syncbn.py will report error. I will try to reimplement the syncbn.py based on v0.4.0 in the future. For now, if you can not install PyTorch from source succussfully, you could use the branch 'pytorch0.4.0' of this project which uncomments the encoding.py and syncbn.py

2. git clone https://github.com/wenqingchu/PyTorch-Encoding

3. Since I use anaconda3, I install this project by "python setup.py install --prefix=~/anaconda3/"

Train

1. There are fcn.py, deeplab.py, psp.py and encnet.py model in encoding/model/.
2. If you want to train on your own dataset, you should implement a dataset file like encoding/datasets/cityscapes.py.
3. Please read experiment/segmentation/option.py to adjust the hyper parameter setting. I use batch_size=2, lr=0.003, 2 gpus, psp model, resnet101, 80 epoches, and the model can achieve mIoU=0.75 on cityscapes validation dataset.
4. Here is an example for train psp model on cityscapes. CUDA_VISIBLE_DEVICES=0,1,2,3 python train_cityscapes.py --dataset cityscapes --model psp --backbone resnet101 --aux

Test

CUDA_VISIBLE_DEVICES=0,1,2,3 python test_cityscapes.py --dataset cityscapes --model psp --backbone resnet101 --aux --eval --resume=runs/cityscapes/psp/default/checkpoint.pth.tar

Citations

Context Encoding for Semantic Segmentation [arXiv]
Hang Zhang, Kristin Dana, Jianping Shi, Zhongyue Zhang, Xiaogang Wang, Ambrish Tyagi, Amit Agrawal

@InProceedings{Zhang_2018_CVPR,
author = {Zhang, Hang and Dana, Kristin and Shi, Jianping and Zhang, Zhongyue and Wang, Xiaogang and Tyagi, Ambrish and Agrawal, Amit},
title = {Context Encoding for Semantic Segmentation},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2018}
}

Deep TEN: Texture Encoding Network [arXiv]
Hang Zhang, Jia Xue, Kristin Dana

@InProceedings{Zhang_2017_CVPR,
author = {Zhang, Hang and Xue, Jia and Dana, Kristin},
title = {Deep TEN: Texture Encoding Network},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {July},
year = {2017}
}