| 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.
The goal is that from a video, we extract the background
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+ |
 |
|
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
http://www.pyimagesearch.com/2015/06/22/install-opencv-3-0-and-python-2-7-on-ubuntu/