NoisyIC

Official Pytorch Implementation for

"Learning-based Compression for Noisy Images in the Wild"

Installation

git clone https://github.com/ppingzhang/NoisyIC.git

conda create -n NoisyIC python=3.8 conda activate NoisyIC pip install -r requirement.txt

Note: torch, GPU and CUDA version need match!

Dataset

How to obtain the dataset:

Flickr3k: download1 or download2

SSID: download

the structure of image path
├── flickr3k
│   ├── 9964318083_6199f7ee20_b.jpg
│   ├── 9964619774_c10a0480df_b.jpg
│   ├── ...
├── SSID
│   ├── img0000.png
│   ├── img0000_gt.png
│   ├── img0001.png
│   ├── img0001_gt.png
├── ...

**Note: Your can originze your own dataset following above data structure. ** Or you can create your own rule to find the imags in the "data/dataset_load.py"

Run the training code

• VGG19 pretrain model need to be downloaded link and please set it in the ./loss file

• Pretrained model can be download through this link

python main.py --mode=train --train_img_dataset='./dataset/flickr3k' --train_real_dataset='./dataset/SSID' --model=MainCodec --lmbda=1 #[1, 5, 20, 50]

Run the testing code

python main.py --mode=test --model=MainCodec --test_dataset_de=./noisy_images_path/ --test_dataset_gt=./clean_images_path/ --ckpt=''

Test a single image

python demo.py --ckpt=test --img_path="./ckpt.pth.tar" --img_save_path="./xx_decode.jpg" --bin_save_path="./xx_bitstream.bin"

BibTeX

@ARTICLE{Learning2022Zhang,
  author={Zhang, Pingping and Wang, Meng and Chen, Baoliang and Lin, Rongqun and Wang, Xu and Wang, Shiqi and Kwong, Sam},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={Learning-based Compression for Noisy Images in the Wild}, 
  year={2022},
  volume={},
  number={},
  pages={1-1}}