santonugoswami / arcticnet

[CVPRW 2019] ArcticNet: A Deep Learning Solution to Classify Arctic Wetlands

ArcticNet: A Deep Learning Solution to Classify Arctic Wetlands

This repository contains the code and trained models for the work:

Ziyu Jiang Kate Von Ness Julie Loisel Zhangyang Wang
ArcticNet: A Deep Learning Solution to Classify Arctic Wetlands 
Published at CVPR 2019 Detecting Objects in Aerial Images Workshop

If you find the code useful for your work, please consider citing

@article{jiang2019arcticnet,
  title={ArcticNet: A Deep Learning Solution to Classify Arctic Wetlands},
  author={Jiang, Ziyu and Von Ness, Kate and Loisel, Julie and Wang, Zhangyang},
  journal={arXiv preprint arXiv:1906.00133},
  year={2019}
}

Our dataset and trained model can be found here.

Prerequisite

This code requires Pytorch 1.0.0, we only support single GPU.

To install Pytorch, please refer to the docs at the Pytorch website.

Other requirements contains numpy, cv2, pyproj, shapefile, gdal, random, pandas.

For installing gdal, you can consider conda install -c conda-forge gdal

Data Preparation

For running the model, you need first download the data.

Then you need to extract it and put the path of it in config file. Specifically, change the value after ROOT: with the /path/to/data/ in config/config_fuse.yaml and config/config_singleBranch.yaml.

Testing

For testing, first download the pretrained models. Create a folder named checkpoint and extract all the pretrained models into it. Form it as

ArcticNet
|--checkpoints
   |--rgbBranch
   |--ndnBranch
   |--fusenet_midFuse_layer3
   |--...
|--config
|--data
...

If testing single branch model. Change the EXPERIENT in config/config_singleBranch.yaml into the folder name of pretrained model and set TESTON as True. Then run:

python train_singleBranch.py

If testing fuse model. Change the EXPERIENT in config/config_fuse.yaml into the folder name of pretrained model and set TESTON as True. Then run:

python train_fuse.py

Training

If train single branch model. Change the EXPERIENT in config/config_singleBranch.yaml into the name you like and set TESTON, MAPGENEON as False. Then run:

python train_singleBranch.py

If train fuse model. Change the EXPERIENT in config/config_fuse.yaml into the name you like and set TESTON, MAPGENEON as False. Then run:

python train_fuse.py

Semantic Map Generating

If using single branch model. Change the EXPERIENT in config/config_singleBranch.yaml into pretrained model name you want to use and set TESTON as False, MAPGENEON as True. Then run:

python train_singleBranch.py

If using fuse model. Change the EXPERIENT in config/config_fuse.yaml into pretrained model name you want to use and set TESTON as False, MAPGENEON as True. Then run:

python train_fuse.py