Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/LavieC/VideoImagination

Last synced: 4 months ago
JSON representation

Host: GitHub
URL: https://github.com/LavieC/VideoImagination
Owner: LavieC
Created: 2017-04-20T16:00:22.000Z (almost 8 years ago)
Default Branch: master
Last Pushed: 2018-02-25T14:18:51.000Z (almost 7 years ago)
Last Synced: 2024-08-01T19:51:32.512Z (7 months ago)
Language: Python
Size: 4.81 MB
Stars: 100
Watchers: 2
Forks: 19
Open Issues: 5
Metadata Files:
- Readme: README.md

Awesome Lists containing this project

README

# Video Imagination from a Single Image with Transformation Generation
This repository contains an implementation of [Video Imagination from a Single Image with Transformation Generation](https://dl.acm.org/citation.cfm?id=3126737). The framework can synthesize multiple imaginary video from a single image.

## Imaginary Video Example
We randomly pick some imaginary videos synthesized by our framework. The input is a single image from UCF101 dataset, and the output imaginary video contains five frames.
The following gif picture is a demo of synthesized imaginary video. The network may bring some delay, please wait a while fro the demonstration.
> Imaginary Video
>
> ![gif](https://github.com/gitpub327/VideoImagination/blob/master/Imaginary_Video.gif)

> Input image
>
>![im](https://github.com/gitpub327/VideoImagination/blob/master/Example_of_Input_Image.png)

## Data
The framework can be trained on three datasets : moving MNIST, 2D shape, UCF101. No pre-process is needed except normalizing images to be in the range [0, 1].
The videos (or image tuples) needs to be convert to tfrecords at first.
## Training
The code requires a TensorFlowr r1.0 installation

To train the framework, after you prepare the tfrecords, run main.py. This file will build model and graph, and train the networks.

## Notes
The code is modified based on [A Tensorflow Implementation of DCGAN](https://github.com/bamos/dcgan-completion.tensorflow). The on-the-fly 2D shape dataset generating codes are modified from [the author of the dataset](https://github.com/tensorflow/models/tree/master/next_frame_prediction).