MTKD

This code is built based on BasicSR, where more information can be found to help use this code.

Thanks to pytorch_wavelets for implementing Wavelet Transforms in Pytorch.

Installation

Please refer to BasicSR

Preparing datasets

Training sets:

[DIV2K] More details are in DatasetPreparation.md

Test sets:

[Set5]

[Set14]

[BSD100]

[urban100]

Model ZOO

EDSR Official Models

SwinIR Official Models

RCAN Official Models

Our trained model will be published soon!!

Use

Step 1 - Train DCTSwin

python .\basicsr\train.py -opt .\options\train\ECCVMTKD\train_DCTSwinx4_3mt.yml

Step 2 - Train EDSRbl

python .\basicsr\train.py -opt .\options\train\ECCVMTKD\train_EDSR_Mx4_KD.yml

Citation

@inproceedings{jiang2024mtkd,
  title={MTKD: Multi-Teacher Knowledge Distillation for Image Super-Resolution},
  author={Jiang, Yuxuan and Feng, Chen and Zhang, Fan and Bull, David},
  booktitle={European Conference on Computer Vision},
  year={2024}
}