https://github.com/GV1028/videogan

Implementation of "Generating Videos with Scene Dynamics" in Tensorflow
https://github.com/GV1028/videogan

generative-adversarial-network tensorflow video video-generation video-representation-learning

Last synced: 11 months ago
JSON representation

Implementation of "Generating Videos with Scene Dynamics" in Tensorflow

Host: GitHub
URL: https://github.com/GV1028/videogan
Owner: GV1028
Created: 2018-03-02T00:59:51.000Z (over 8 years ago)
Default Branch: master
Last Pushed: 2018-03-13T05:04:20.000Z (over 8 years ago)
Last Synced: 2024-11-27T04:30:49.941Z (over 1 year ago)
Topics: generative-adversarial-network, tensorflow, video, video-generation, video-representation-learning
Language: Python
Size: 731 KB
Stars: 77
Watchers: 3
Forks: 20
Open Issues: 8
Metadata Files:
- Readme: README.md

Awesome Lists containing this project

README

# Generating Videos with Scene Dynamics

## Introduction
This repository contains an implementation of "Generating Videos with Scene Dynamics" in Tensorflow. The paper can be found here (http://carlvondrick.com/tinyvideo/paper.pdf). The model learns to generate a video by upsampling from some latent space, using adversarial training.

## Requirements
For running this code and reproducing the results, you need the following packages. Python 2.7 has been used.

Packages:
* TensorFlow
* NumPy
* cv2
* scikit-video
* scikit-image

## VideoGAN - Architecture and Working
Attached below is the architecture used in the paper [paper](http://carlvondrick.com/tinyvideo/paper.pdf).

![Video_GAN](images/videogan.png)

## Usage
Place the videos inside a folder called "trainvideos".

Run main.py with the required values for each flag variable.

## Results
Below are some of the results on the model trained on MPII Cooking Activities dataset.

Real videos