/ImageEnhancement

Camera Image Quality Enhancement AI Contest - DACON Ranked #7

Primary LanguagePythonOtherNOASSERTION

ImageEnhancement

Camera Image Quality Enhancement AI Contest - DACON Ranked #7

This project is based on megvii-model's HINet implementation.

Installation

git clone https://github.com/NextLevel-AI/CameraImageEnhancement
cd CameraImageEnhancement
pip install -r requirements.txt
python setup.py develop --no_cuda_ext

environment

  • Ubuntu 16.04
  • GPU : NVIDIA Tesla V100 x 2
  • Cuda 10.1
  • python 3.7
  • torch 1.7.1
  • cudnn 7.6
  • addict 2.4.0
  • future 0.18.2
  • lmdb 1.2.1
  • numpy 1.19.5
  • opencv-python 4.5.3.56
  • Pillow 8.3.1
  • pyyaml 5.4.1
  • requests 2.26.0
  • scikit-image 0.17.2
  • scipy 1.5.4
  • tb-nightly 2.6.0a20210728
  • tqdm 4.61.2
  • yapf 0.31.0

Pre-processing

mkdir datasets
python scripts/split.py -train [original train_input_img path] -label [original train_label_img path]
python scripts/preprocessing.py

Data augmentation

(Data augmentation code is included during training process. This code is for saving some sample augmented images)

python scripts/augmentation.py

Training

python basicsr/train.py -opt options/train/LG_HINet_Crop512_Step384.yml

Test

python basicsr/inference.py -opt options/inference/HINet-Inference.yml

Weight path

weights/HINet-136k.pth

Result

result1

result2