This is a well written paper, and most concerns were addressed in the well written rebuttal. The main remaining suggestions are perhaps to add another realistic dataset for more complex experiments as well as more analysis is needed on GT class distribution + InfoGAN not performing well.