zjh1124 / deepvideodeblurring

S. Su, M. Delbracio, J. Wang, G. Sapiro, W. Heidrich, O. Wang. Deep Video Deblurring. CVPR 2017, Spotlight

Deep Video Deblurring for Hand-held Cameras

This is the demo code for Deep Video Deblurring for Hand-held Cameras. Given a stack of pre-aligned input frames, our network predicts a sharper central image.

Prepare data

• Download and unzip test videos to dataset, from this link, or place your own test video frames under dataset/qualitative_datasets/[video_file_name]/input.

• Align frames in Matlab, by running

	preprocess/launcher.m

Outputs should be stored at data/testing_real_all_nostab_[alignment] under structure

	/image_-2
	/image_-1
	/image_0
	/image_1
	/image_2

Alternatively, you can download the pre-aligned qualitative videos from here, here, and here.

Download pretrained weights

• Download and unzip pretrained weights into logs, from here.

Run prediction script

• Run script: sh run_pred.sh
• Results will be saved to outImg.

FAQ

• split of the training/test sets: shuochsu#2 (comment)

Citation

If you find this code useful for your research, please cite:

@inproceedings{su2017deep,
  title={Deep Video Deblurring for Hand-held Cameras},
  author={Su, Shuochen and Delbracio, Mauricio and Wang, Jue and Sapiro, Guillermo and Heidrich, Wolfgang and Wang, Oliver},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={1279--1288},
  year={2017}
}