Pytorch implementation of MHCL

Preparation

Install PyTorch following the official PyTorch ImageNet training code.

Unsupervised Training

This implementation only supports multi-gpu, DistributedDataParallel training, which is faster and simpler; single-gpu or DataParallel training is not supported.

To do unsupervised pre-training of a ResNet-50 model on ImageNet in an 4-gpu machine, run:

python main_mhcl.py \
  -a resnet50 \
  --lr 0.015 \
  --batch-size 128 \
  --epochs 800 \
  --mlp --moco-t 0.2 --aug-plus --cos \
  --mhcl --a 1.0 --b 0.1 \
  --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0 \
  [your mlrsnet(4:6)-folder with train and val folders]

For UC Merced 6:4, use --a 20.0 --b 0.1; For NWPU-RESISC45 2:8, use --a 10.0 --b 0.1; For NWPU-RESISC45 8:2, use --a 10.0 --b 1.0

Linear Probe

With a pre-trained model, to train a supervised linear classifier on frozen features/weights in an 4-gpu machine, run:

python main_lincls.py \
  -a resnet50 \
  --lr 0.1 \
  --batch-size 256 \
  --pretrained [your checkpoint path]/checkpoint_0799.pth.tar \
  --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0 \
  [your mlrsnet-folder with train and val folders]

Citation

Please cite this paper if it helps your research:

@article{li2023contrastive,
  title={Contrastive learning based on multiscale hard features for remote-sensing image scene classification},
  author={Li, Zhihao and Hou, Biao and Guo, Xianpeng and Ma, Siteng and Cui, Yanyu and Wang, Shuang and Jiao, Licheng},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  volume={61},
  pages={1--13},
  year={2023},
  publisher={IEEE}
}

Acknowledgments

This code is built using the moco repository.