Reverse-Matchmove-GAN

The goal of this project is to be able to reverse engineer new camera views from existing footage. This is the same network from a series of blog posts I'm writing:

http://neuralvfx.com/matchmove/reverse-matchmove-gan/

http://neuralvfx.com/matchmove/reverse-matchmove-discriminator/

My dataset was created by photographing many random angles of a statue in Chiang Mai. Using a photo-modeling tool I've extracted camera positions and created a CSV file of Matrix-Image pairs.

The Chiang Mai dataset can be downloaded here: http://neuralvfx.com/datasets/reverse_matchmove/chiang_mai_hi.rar

Generated Video Example

Code Usage

Usage instructions found here: user manual page.

Example Data Set

Example Results