/PBRID

MegCup 2022, blind raw image denoising, 10th place

Primary LanguagePython

A Light-weight model for Practical Blind Raw Image Denoising (PBRID)

visitors

A baseline light-weight model for practical blind raw image denoising (PBRID)

Raw (RGGB) image denoising results on test set:
model Param inference speed (img 256x256, single V100)
97.2K 10.1ms

Dataset & pretrained model

## download dataset
wget -nc https://rutgers.box.com/shared/static/tx3s87qdcx3g8vc62ukf4hk56h0fwn2f.zip -O burst_raw.zip
unzip burst_raw.zip
## download pretrained model file (includes model, ema model and optimizer)
wget -nc https://rutgers.box.com/shared/static/5cluf6gdevwsi7samjkytlt15f3zziaa.pth -P weight/
mv weight/5cluf6gdevwsi7samjkytlt15f3zziaa.pth weight/hqs.pth

Run scripts

## train:
CUDA_VISIBLE_DEVICES=0 python main.py --mode 'train'
## val:
CUDA_VISIBLE_DEVICES=1 python main.py --mode 'val'
## test:
CUDA_VISIBLE_DEVICES=1 python main.py --mode 'test'