PVDD

This is an official implementation of the paper "Towards Real-World Video Denosing: A Practical Video Denosing Dataset and Network". [PDF]

Code

Dependencies

Python 3.6
PyTorch >= 1.1.0
numpy
cv2
skimage
DCNv2
easydict
yaml

Quick Start

Clone this github repo.

git clone https://github.com/Marinyyt/PVDD.git
cd PVDD

Training

Download PVDD|CRVD|DAVIS dataset and unpack them to any place you want.
Run train.py using the corresponding yaml files. (Please change the data_path argument in yaml files and noise-level file path in Dataset class.)

# PVDD sRGB 
python train.py --config /USER_PATH/PVDD/configs/PVDD_pvdd0815_02_charbo_bs1_pvdd_model.yaml --save_path /USER_SAVE_PATH
python train.py --config /USER_PATH/PVDD/configs/PVDD_pvdd0815_02_level_charbo_bs1_pvdd_model.yaml --save_path /USER_SAVE_PATH

# PVDD RAW
python train.py --config /USER_PATH/PVDD/configs/PVDD_pvdd0815_02_charbo_bs1_pvdd_raw_model.yaml --save_path /USER_SAVE_PATH
python train.py --config /USER_PATH/PVDD/configs/PVDD_pvdd0815_02_level_charbo_bs1_pvdd_raw_model.yaml --save_path /USER_SAVE_PATH

# CRVD sRGB
python train.py --config /USER_PATH/PVDD/configs/PVDD_pvdd0815_charbo_bs1_crvd_model.yaml --save_path /USER_SAVE_PATH

# DAVIS sRGB
python train.py --config /USER_PATH/PVDD/configs/PVDD_pvdd0815_charbo_bs1_davis_model.yaml --save_path /USER_SAVE_PATH

You can find the results and logs in save_path.

Testing

Download our pre-trained models and unpack them to any place you want or use your pre-trained models.
Run.

# PVDD
python test_video_pvdd_server.py --model_file /USER_MODEL_CKPT_PATH --save_path /USER_SAVE_PATH --test_path /USER_TEST_DATA_PATH --num_frame 5
python test_video_pvdd_level_server.py --model_file /USER_MODEL_CKPT_PATH --save_path /USER_SAVE_PATH --test_path /USER_TEST_DATA_PATH --num_frame 5

python test_video_pvdd_raw_server.py --model_file /USER_MODEL_CKPT_PATH --save_path /USER_SAVE_PATH --test_path /USER_TEST_DATA_PATH --num_frame 5
python test_video_pvdd_level_raw_server.py --model_file /USER_MODEL_CKPT_PATH --save_path /USER_SAVE_PATH --test_path /USER_TEST_DATA_PATH --num_frame 5

# DAVIS
python test_video_davis_server.py --model_file /USER_MODEL_CKPT_PATH --save_path /USER_SAVE_PATH --test_path /USER_TEST_DATA_PATH --num_frame 5

# CRVD
python test_video_crvd_server.py --model_file /USER_MODEL_CKPT_PATH --save_path /USER_SAVE_PATH --test_path /USER_TEST_DATA_PATH --num_frame 5

pre-trained models

Google Drive | Baidu Drive

Dataset

Please download PVDD from Google Drive or Baidu Drive.

	sRGB	raw
Training Dataset	Google Drive, Baidu Drive	Google Drive, Baidu Drive
Testing Dataset	Google Drive, Baiidu Drive	Google Drive, Baidu Drive

Citation

If you find our work useful in your research or publication, please cite:

@article{xu2022pvdd,
  title={Pvdd: A practical video denoising dataset with real-world dynamic scenes},
  author={Xu, Xiaogang and Yu, Yitong and Jiang, Nianjuan and Lu, Jiangbo and Yu, Bei and Jia, Jiaya},
  journal={arXiv preprint arXiv:2207.01356},
  year={2022}
}

Marinyyt / PVDD