English | 简体中文

This is the pytorch implement of our paper "Continuous Remote Sensing Image Super-Resolution based on Context Interaction in Implicit Function Space"

Project Page $\cdot$ PDF Download $\cdot$ HuggingFace Demo

conda create -n FunSR python=3.10

conda activate FunSR

Version of 1.x is also work, but the version of 2.x is recommended.

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu117

conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia

Version of 2.x is recommended, but the version of 1.x is also work.

pip install mmcv==2.0.0 -f https://download.openmmlab.com/mmcv/dist/cu117/torch2.0/index.html

Please refer to installation documentation for more detailed installation.

pip install -r requirements.txt

Put the downloaded HR images into the samples folder. In this project, some example images are provided in this folder.

The data split files in the paper are provided in the data_split folder; if you need to split the training and validation set by yourself, please use tools/data_tools/get_train_val_list.py to split the training and validation set, run python tools/data_tools/get_train_val_list.py.

The config file of FunSR is configs/train_1x-5x_INR_funsr.yaml. You can modify the parameters in this file according to the situation.

Run python train_inr_funsr.py to train the FunSR model. And you can modify the ArgumentParser parameters in this file according to the situation.

The config file of fixed-scale SR models is configs/baselines/train_CNN.yaml.

Run python train_cnn_sr.py to train the fixed-scale SR models.

The config file of continuous-scale SR models is configs/baselines/train_1x-5x_INR_[liif, metasr, aliif].yaml.

Run python train_liif_metasr_aliff.py to train the continuous-scale SR models.

The config file of continuous-scale SR models is configs/baselines/train_1x-5x_INR_diinn_arbrcan_sadn_overnet.yaml.

Run python train_diinn_arbrcan_sadn_overnet.py to train the continuous-scale SR models.

The config file of FunSR is configs/test_INR_diinn_arbrcan_funsr_overnet.yaml. You can modify the parameters in this file according to the situation.

Run python test_inr_diinn_arbrcan_sadnarc_funsr_overnet.py to test the FunSR model. And you can modify the ArgumentParser parameters in this file according to the situation.

The config file of interpolation-based SR models is configs/test_interpolate.yaml.

Run python test_interpolate_sr.py to test the interpolation-based SR models.

The config file of fixed-scale SR models is configs/baselines/test_CNN.yaml.

Run python test_cnn_sr.py to test the fixed-scale SR models.

The config file of continuous-scale SR models is configs/baselines/test_1x-5x_INR_[liif, metasr, aliif].yaml.

Run python test_liif_metasr_aliff.py to test the continuous-scale SR models.

The config file of continuous-scale SR models is configs/baselines/test_1x-5x_INR_diinn_arbrcan_sadn_overnet.yaml.

Run python test_diinn_arbrcan_sadn_overnet.py to test the continuous-scale SR models.

In order to evaluate the models of different super-resolution ratios conveniently, we provide a batch evaluation script, which is located in scripts/test_script.py, which can be run python scripts/test_script.py

Some visualization tools are provided in the tools/paper_vis_tools folder, you can refer to the files in this folder for details.

The model weights of RDN are provided in the huggingface space.

If you find this project useful for your research, please cite our paper.

If you have any other questions, please contact me!!!

@article{chen2023continuous,
  title={Continuous Remote Sensing Image Super-Resolution based on Context Interaction in Implicit Function Space},
  author={Chen, Keyan and Li, Wenyuan and Lei, Sen and Chen, Jianqi and Jiang, Xiaolong and Zou, Zhengxia and Shi, Zhenwei},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  year={2023},
  publisher={IEEE}
}