Awesome

ERDN: Equivalent Receptive Field Deformable Network for Video Deblurring (ECCV 2022)

[[Paper]]

Bangrui Jiang, Zhihuai Xie, Zhen Xia, Songnan Li, Shan Liu

Tencent Media Lab, Shenzhen, China

Dependencies

• Python >= 3.6 (Recommend to use Anaconda or Miniconda)
• PyTorch >= 1.1
• NVIDIA GPU + CUDA
• Linux

Installation

Clone repo

git clone xxx
cd ERDN

Install dependent packages

• torchvision
• tqdm
• imageio
• numpy
• opencv-python

We use Deformable-Convolution-v2 and install as follows

cd dcn
bash make.sh
cd ..

Data Preparation

We use DVD for training and testing. The dataset can be download as follows

wget http://www.cs.ubc.ca/labs/imager/tr/2017/DeepVideoDeblurring/DeepVideoDeblurring_Dataset.zip
unzip DeepVideoDeblurring_Dataset.zip

The data should be placed according to the following format

|--DVD
    |--Train
        |--blur 
            |--video 1
                |--frame 1
                |--frame 2
                    :  
            |--video 2
                :
            |--video n
        |--gt
            |--video 1
                |--frame 1
                |--frame 2
                    :
            |--video 2
                :
            |--video n
    |--Test
        |--blur
            |--video 1
                :
        |--gt
            |--video 1
                :

We provide preprocess script for DVD dataset

python script/arrange.py --data_path path_to_origin_DVD_dataset --out_path path_to_DVD

Quick Inference

Download pre-trained models from DVD (key: csig).

Run following command for quick inference.

python inference.py \
--data_path path_to_DVD \
--model_path path_to_model \
--result_path path_to_save_result \ 
--save_image whether_to_save_image

Training

The training script will be released soon.