Home

Awesome

MegCup 2022 Team Feedback

This repository is the 1st place solution (Team Feedback) in 2022 MegCup RAW image denoising. Here is the 3rd place solution, FeedBack-based Restormer.

Members

Overview

<div align="center"> <img src="figs/fig-1.jpg" width="782" height="435"> </div> <details><summary>Click for more illustration</summary> <div align="center"> <img src="figs/fig-2.jpg" width="782" height="435"> <img src="figs/fig-3.jpg" width="782" height="435"> <img src="figs/fig-4.jpg" width="782" height="435"> </div> </details>

Environment

git clone https://github.com/hlh981029/megcup-feedback.git
cd megcup-feedback

conda env create -f environment.yaml
conda activate feedback

Dataset

Please download the dataset to ./data, and refer to options/feedback.yaml to modify the data path.

|--data
   |--competition_train_input.0.2.bin
   |--competition_train_gt.0.2.bin
   |--competition_test_input.0.2.bin

Evaluation

# evaluate on dataset
# log and config file will be saved to ./output/feedback

python test.py

# generate result bin file
# result will be saved to ./output/feedback/submit/model_best_result.bin

python test.py --submit

Acknowledgement

This project is based on Global2Local, Swin-Transformer, Restormer, and BasicSR.