Variational autoencoders & unets: resources and discussion

Variational autoencoders (VAEs) and unets are an important component of Stable Diffusion. Use this thread for asking questions, sharing links, or chatting about stuff related to these topics.

Here’s a post about VAEs from @irhumshafkat: https://towardsdatascience.com/intuitively-understanding-variational-autoencoders-1bfe67eb5daf

38 Likes

I was watching some youtube videos like this one and still having quite a vague understanding, but @irhumshafkat 's blog gave a very intuitive explanation about the reconstruction loss and regularization loss.

5 Likes

Thanks for sharing this. It’s an amazingly clear and well written post

4 Likes

IIRC he was still at high school when he wrote this! :smiley:

10 Likes

that is amazing.

This is a good video for intuition in my opinion: Variational Autoencoders - EXPLAINED by CodeEmporium

3 Likes

Also a good blog post that explains Variational Autoencoders vs GANs in an intuitive way

Variational Autoencoders Explained by Kevin Frans 2016.

2 Likes

I found this really accessible and detailed. An Introduction to VAEs

1 Like

A very detailed, yet mostly informal introduction to Encoder-Decoder architectures: Understanding Variational Autoencoders (VAEs) | by Joseph Rocca | Towards Data Science.

The article is quite long (by Medium standards), and with good reasons.

Rocca is, in general, one of my favourite authors.

2 Likes