PFSegNets

This repo contains the the implementation of Our CVPR-2021 work: PointFlow: Flowing Semantics Through Points for Aerial Image Segmentation

PFSegNet implemented by Jittor from THU.

To do

Support PointFlow Models
Add more Models (at least 15 models)

The master branch works with PyTorch 1.5 and python 3.7.6.

DataSet preparation

Downloading iSAID, Potsdam and Vahihigen dataset.
Using scripts to crop iSAID and Potsdam, Vaihigen into patches.
Using scripts to convert the original mask of iSAID and Potsdam, Vaihigen into gray mask for training and evaluating.
Finally, you can either change the config.py or do the soft link according to the default path in config.

For example, suppose you store your iSAID dataset at ~/username/data/iSAID, please update the dataset path in config.py,

__C.DATASET.iSAID_DIR = '~/username/data/iSAID'

Or, you can link the data path into current folder.

mkdir data 
cd data
ln -s your_iSAID_root_data_path iSAID

Install the extension for other models

# install the extension 
python setup.py develop

Actually, the order of steps 2 and 3 is interchangeable.

Pretrained Models

Baidu Pan Link: https://pan.baidu.com/s/1MWzpkI3PwtnEl1LSOyLrLw 4lwf

Google Drive Link: https://drive.google.com/drive/folders/1C7YESlSnqeoJiR8DWpmD4EVWvwf9rreB?usp=sharing

After downloading the pretrained ResNet, you can either change the model path of network/resnet_d.py or do the soft link according to the default path in network/resnet_d.py.

For example, Suppose you store the pretrained ResNet50 model at ~/username/pretrained_model/resnet50-deep.pth, please update the dataset path in Line315 of config.py,

model.load_state_dict(torch.load("~/username/pretrained_model/resnet50-deep.pth", map_location='cpu'))

Or, you can link the pretrained model path into current folder.

mkdir pretrained_models
ln -s your_pretrained_model_path path_to_pretrained_models_folder

Model Checkpoints

Dataset	Backbone	mIoU	Model
iSAID	ResNet50	66.9	Google Drive \| Baidu Pan(v3oj)
Potsdam	ResNet50	75.4	Google Drive \| Baidu Pan(lhlf)
Vaihigen	ResNet50	70.4	Google Drive \| Baidu Pan(54qm)

Evaluation

For example, when evaluating PFNet on validation set of iSAID dasaset:

sh scripts/pointflow/test/test_iSAID_pfnet_R50.pth path_to_checkpoint path_to_save_results

If you want to save images during evaluating, add args: dump_images, which will take more time.

Training

To be note that, our models are trained on 8 V-100 GPUs with 32GB memory. It is hard to reproduce such best results if you do not have such resources. For example, when training PFNet on iSAID dataset:

sh scripts/pointflow/train_iSAID_pfnet_r50.sh

Citation

If you find this repo is helpful to your research. Please consider cite our work.

@inproceedings{li2021pointflow,
  title={PointFlow: Flowing Semantics Through Points for Aerial Image Segmentation},
  author={Li, Xiangtai and He, Hao and Li, Xia and Li, Duo and Cheng, Guangliang and Shi, Jianping and Weng, Lubin and Tong, Yunhai and Lin, Zhouchen},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={4217--4226},
  year={2021}
}

Acknowledgement

This repo is based on NVIDIA segmentation repo. We fully thank their open-sourced code.

lxtGH/PFSegNets