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).
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
Please refer to the paper paper.
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
.
Run test_golf.py
.
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![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |