git.sesse.net Git - voxel-flow/blob - README.md

   1 # Video Frame Synthesis using Deep Voxel Flow
   2 We address the problem of synthesizing new video frames in an existing video, either in-between existing frames (interpolation), or subsequent to them (extrapolation). Our method requires no human supervision, and any video can be used as training data by dropping, and then learning to predict, existing frames. `Deep Voxel Flow (DVF)` is efficient, and can be applied at any video resolution. We demonstrate that our method produces results that both quantitatively and qualitatively improve upon the state-of-the-art.
   3
   4 [[Project]](https://liuziwei7.github.io/projects/VoxelFlow) [[Paper]](https://arxiv.org/abs/1702.02463) [[Demo]](https://liuziwei7.github.io/projects/voxelflow/demo.html)
   5
   6 <img src='./misc/demo.gif' width=810>
   7
   8 ## Overview
   9 `Deep Voxel Flow (DVF)` is the author's re-implementation of the video frame synthesizer described in:
  10 "Video Frame Synthesis using Deep Voxel Flow"
  11 [Ziwei Liu](https://liuziwei7.github.io/), [Raymond A. Yeh](http://www.isle.illinois.edu/~yeh17/), [Xiaoou Tang](http://www.ie.cuhk.edu.hk/people/xotang.shtml), [Yiming Liu](http://bitstream9.me/), [Aseem Agarwala](http://www.agarwala.org/) (CUHK & UIUC & Google Research)
  12 in International Conference on Computer Vision (ICCV) 2017, Oral Presentation
  13
  14 <img src='./misc/demo_teaser.jpg' width=800>
  15
  16 Further information please contact [Ziwei Liu](https://liuziwei7.github.io/).
  17
  18 ## Requirements
  19 * [TensorFlow](https://www.tensorflow.org/)
  20
  21 ## Getting started
  22 * Run the training script:
  23 ``` bash
  24 python voxel_flow_train.py --subset=train
  25 ```
  26 * Run the testing script:
  27 ``` bash
  28 python voxel_flow_train.py --subset=test
  29 ```
  30 * Run the evaluation script:
  31 ``` bash
  32 matlab eval_voxelflow.m
  33 ```
  34
  35 ## License and Citation
  36 The use of this software is RESTRICTED to **non-commercial research and educational purposes**.
  37
  38 ```
  39 @inproceedings{liu2017voxelflow,
  40  author = {Ziwei Liu, Raymond Yeh, Xiaoou Tang, Yiming Liu, and Aseem Agarwala},
  41  title = {Video Frame Synthesis using Deep Voxel Flow},
  42  booktitle = {Proceedings of International Conference on Computer Vision (ICCV)},
  43  month = {October},
  44  year = {2017}
  45 }
  46 ```
  47
  48 ## Disclaimer
  49 This is not an official Google product.