Dref360 / bg_estimation

Background estimation using deeplearning

Background estimation using deep learning (Not maintained)

Input	Output

This repo has 4 models to estimate the background of a video.

VGG3D, UNET, VAE (not working), SingleFrame (not using video)

The database used is from : http://scenebackgroundmodeling.net/

I recommend being on Keras master branch since there is a lot of fixes.

This repo is using Tensorflow, it would not works using Theano since the SSIM is not implemented using the Keras Backend.

Goal

The goal is that from a video, we extract the background

VGG+

I tested many models, but the best seems to be VGG+, a VGG3D implementation with a skip-connection.

For the loss, I used SSIM with an MSE.

Here's a comparison between VGG3D and VGG+

VGG3D	VGG+

Examples

Install Tensorflow

Refer to this page to be kept up to date! https://www.tensorflow.org/versions/r0.11/get_started/os_setup.html#virtualenv-installation This repo is using tf 0.11.0rc

Install cv3 on Ubuntu

http://www.pyimagesearch.com/2015/06/22/install-opencv-3-0-and-python-2-7-on-ubuntu/