/videogan-pytorch

Primary LanguagePython

Generating Videos with Scene Dynamics

Introduction

This is a Pytorch implementation of "Generating Videos with Scene Dynamics" (https://github.com/cvondrick/videogan). Codes are largely from a TensorFlow implementation (https://github.com/GV1028/videogan).

Requirements

This code was tested under Pytorch 0.4.0. Required packages:

  • pip install numpy
  • pip install pillow
  • pip install scikit-image
  • pip install scikit-video

VideoGAN

Please refer to the paper paper.
Video_GAN

Train

Place golf dataset inside a folder golf.
(You can download the dataset from http://www.cs.columbia.edu/~vondrick/tinyvideo/).
Run train_golf.py with the required values in argparser.

Test

Run test_golf.py.

Results

Below are some of the results on the model trained on golf dataset about 3 epochs.
More training will likely yield better results.

Generated videos