Deblur-by-Fitting

We present a novel approach to video deblurring by fitting a deep network to the test video. One key observation is that some frames in a video with motion blur are much sharper than others, and thus we can transfer the texture information in those sharp frames to blurry frames. Our approach heuristically selects sharp frames from a video and then trains a convolutional neural network on these sharp frames. The trained network often absorbs enough details in the scene to perform deblurring on all the video frames. As an internal learning method, our approach has no domain gap between training and test data, which is a problematic issue for existing video deblurring approaches. The conducted experiments on real-world video data show that our model can reconstruct clearer and sharper videos than state-of-the-art video deblurring approaches.

Results on GoPro dataset:
synthetic

Requirement:

Tensorflow 1.x

Set up:

  • $ mkdir VGG_Model
  • Download VGG-19. Search imagenet-vgg-verydeep-19 in this page and download imagenet-vgg-verydeep-19.mat. We need the pre-trained VGG-19 model for our hypercolumn input and feature loss
  • move the downloaded vgg model to folder VGG_Model

Training

  1. Download the dataset and put it under ./dataset

  2. Run

python train.py

Testing

  1. Run
python test.py --output the_output_path

Meta-learning

Dataset

Citation

If you use this code for your research, please cite our paper.

@InProceedings{ren_deblur,
author = {Xuanchi Ren, Zian Qian, Qifeng Chen},
title = {Video Deblurring by Fitting to Test Data},
booktitle = {arxiv},
year = {2020}
}

GitHub