PLKSR: Partial Large Kernel CNNs for Efficient Super-Resolution

August 19, 2024 ยท View on GitHub

image

This repository is an official implementation of the paper "Partial Large Kernel CNNs for Efficient Super-Resolution", Arxiv, 2024.

by Dongheon Lee, Seokju Yun, and Youngmin Ro

[paper] [pretrained models]

Updates

  • [2024-08-19] PLKSR-IGConv+, capable of predicting multiple integer scales with a single model, has been released and is available in the repository. IGConvPlus
  • [2024-05-22] Pre-trained models of the PLKSR on the DF2K dataset are released. df2k_quantitative
  • [2024-05-10] Real-PLKSR, to train PLKSR stably on real-world SISR task, has been provided. Implementation details are available in issue and you train/test it with the neosr framework.

Installation

git clone https://github.com/dslisleedh/PLKSR.git
cd PLKSR
conda create -n plksr python=3.10
conda activate plksr
conda install pytorch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 pytorch-cuda=12.1 -c pytorch -c nvidia
pip install -r requirements.txt
python setup.py develop

Train

python plksr/train.py -opt=$CONFIG_PATH

Test

python plksr/test.py -opt=$CONFIG_PATH

Results

Quantitative Results

Main model

image

Tiny model

image

Visual Results

image image

Acknowledgement

This work is released under the MIT license. The codes are based on BasicSR. Thanks for their awesome works.

Contact

If you have any questions, please contact dslisleedh@gmail.com