z_xh <- mu + epsilon * sigma_0. How is the sigma_0 (0.0032) selected?

#11 · open · 0 comments

View on GitHub ↗

VicFisher

Hello, Minkai: I found in the VAE encoder that you implemented the reparameteraztion process `z_xh <- mu + epsilon * sigma_0` with `sigma_0 = 0.0032`. How to select a proper sigma_0 ? Is there any special skills ?

Comments