vztu / DebandingNet

DebandingNet

Installation

• Install yacs: conda install -c conda-forge yacs
• Install natsort: pip install natsort
• Install skimage: pip install scikit-image
• Install tqdm: pip install tqdm
• Install cv2: conda install -c conda-forge opencv
• Install warmup scheduler

cd pytorch-gradual-warmup-lr; python setup.py install; cd ..

Download dataset

TBD

Start training

• Carefully modify training.yml and then run:

python3 train.py

Testing on validation set

• Download images of validation set and place them in datasets/VP9BandingDataset/val/

• Run:

python test.py --input_dir datasets/ --model_file "UNet.py" --model_variant "UNet-32" --result_dir results/test_figure5_AdaDeband_UNet/ --dataset FFMPEG --weights checkpoints/Debanding/models/UNet32_epochs300_l1loss_train256x256/model_latest.pth --gpus='0,1' --crop_size=0

Acknowledgment

This code is based on MPRNet