Official PyTorch Implementation of the ECCV 2022 Paper
Project | arXiv
This repo contains training and evaluation code for the following paper:
Deep Semantic Statistics Matching (D2SM) Denoising Network
Kangfu Mei, Vishal M. Patel , and Rui Huang
European conference on computer vision (ECCV) 2022
Use D2SM in your project can be done in two commands as
pip install git+https://github.com/MKFMIKU/d2sm.git
from d2sm import DeST
dest = DeST("54", args.patch_size, args.patch_size, Q=args.dest_q, K=args.dest_k, mean_shift=True)
dest_loss = dest(self.hr, self.denoised_hr)
!!!
The code style is based on Black.
First install pytorch
pip install -i https://opentuna.cn/pypi/web/simple torch==1.7.1+cu101 torchvision==0.8.2+cu101 -f https://download.pytorch.org/whl/torch_stable.html
Than install requirements
pip install -r ./requirements.txt
- For D2SM training on FFDNet with CityscapesNoise dataset
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7,8 python train.py --data_root /ram_data/Refined-Cityscapes-Denoising/ --multiprocessing-distributed --world-size 1 --rank 0 --dist-url tcp://localhost:10001 --batch-size 8 --workers 16 --model FFDNet --dataset CityscapesNoise --rgb_range 1 --lr 1e-4 --checkpoint /data/Experiments/PyAnole/DeST --epochs 400 --patch_size 512 --display_freq 5000 --print_freq 10 --test_freq 20 --name FFDNet_InnerDeST --dest
- For D2SM testing on FFDNet with CityscapesNoise dataset
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7,8 python test.py --data_root /ram_data/Refined-Cityscapes-Denoising/val/ --multiprocessing-distributed --world-size 1 --rank 0 --dist-url tcp://localhost:10001 --batch-size 4 --workers 16 --model FFDNet --dataset imgfolder --rgb_range 1 --lr 1e-4 --checkpoint /data/Experiments/PyAnole/DeST --epochs 800 --display_freq 1000 --print_freq 10 --test_freq 10 --name FFDNet_L1 --output /data/Experiments/PyAnole/Results/FFDNet_L1 --resume /data/Experiments/PyAnole/DeST/FFDNet_L1/checkpoint_0400.pth.tar --sigma 25
- Utilized measurement in the paper
python tools/quality_measure.py --dir2 /tmp_data/running/Refined-Cityscapes-Denoising/val/ --dir1 /data/Experiments/TOFNet/Results/FFDNet/inpktvgg_25/output/ --cuda -b 8