Transitional Learning: Exploring the Transition States of Degradation for Blind Super-resolution (TLSR)

This repository is for TLSR introduced in the following paper

Yuanfei Huang, Jie Li, Yanting Hu, Xinbo Gao and Hua Huang*, "Transitional Learning: Exploring the Transition States of Degradation for Blind Super-resolution", IEEE TPAMI, 2023, 45(5): 6495-6510. paper

Dependenices

• python 3.7
• pytorch >= 1.5
• NVIDIA GPU + CUDA

Models

Download the pre-trained models from Google Drive or 百度网盘 (提取码: ohwt)

Data preparing

Download DIV2K and Flickr2K datasets into the path "data/Datasets/Train/DF2K".

Settings (option.py)

For convolutive degradations (isotropic Gaussian):

• '-scale' == 2

• '-degrad_train' == {'type': 'B', 'min_sigma': 0.2, 'max_sigma': 2.0} # for Training.

• '-degrad_test' == [{'type': 'B', 'sigma': 1.0}] # for Testing.

  OR

• '-scale' == 4

• '-degrad_train' == {'type': 'B', 'min_sigma': 0.2, 'max_sigma': 4.0} # for Training.

• '-degrad_test' == [{'type': 'B', 'sigma': 2.0}] # for Testing.

For convolutive degradations (anisotropic Gaussian):

• '-scale' == 4
• '-degrad_train' == {'type': 'B_aniso', 'min_sigma': 0, 'max_sigma': 0.5}
• '-degrad_test' == [{'type': 'B_aniso', 'sigma': 0.25}] # for evaluation.

For additive degradations:

• '-scale' == 1 OR 2 OR 4
• '-degrad_train' == {'type': 'N', 'min_sigma': 0, 'max_sigma': 30}
• '-degrad_test' == [{'type': 'N', 'sigma': 15}] # for evaluation.

For other degradations:

• '-scale' == 1
• '-degrad_train' == {'type': 'JPEG', 'min_sigma': 10, 'max_sigma': 30}
• '-degrad_test' == [{'type': 'JPEG', 'sigma': 20}] # for evaluation.

Train

python main.py --train 'Train'

Test

python main.py --train 'Test'

Citation

@ARTICLE{TLSR2022TPAMI,
  author={Huang, Yuanfei and Li, Jie and Hu, Yanting and Gao, Xinbo and Huang, Hua},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, 
  title={Transitional Learning: Exploring the Transition States of Degradation for Blind Super-Resolution}, 
  year={2023},
  volume={45},
  number={5},
  pages={6495-6510},
  doi={10.1109/TPAMI.2022.3206870}}