ML Background Matting (2020)
Experimentation of background matting ML without green screen
ML Model: Background Matting: The World is Your Green Screen, University of Washington
Environment: TensorFlow 2.0 on Colab
Input video captured using a mobile device (Pixel 3a) in my basement.
Background plate provided in the researcher's GitHub
Composited result showing great potential without green screen required
The result shows great potential that with mobile device and without laborious green screen setup, it is possible to perform the background matting for video production. The matting is not clean since the footage was filmed in the basement. The close proximity to the backwall probably causes some self-shadowing. I am excited to experiment more to verify the theory with outdoor footage.