entagma
patreon
Advanced Setups – pt.27: PyTorch Series 2a – Upgrading the Autoencoder
🕑 Added 2024-01-11 14:16:51 +0000 UTCChris has been busy with PyTorch again! In the next two videos we will continue where we left off with our MNIST autoencoder in the last Advanced Setups video. In part 2a we will upgrade our autoencoder to process larger color images and feed it two custom datasets – One that works pretty well and one that will show us the limits of our network architecture. In part 2b we will learn how to create 3d maps of our latent spaces, even if those spaces themselves have a couple hundred dimensions. We will also compare the latent space of our own custom net to OpenAI’s Clip model as well as some good old fashioned algorithms to see, who can represent our datasets the best.
One sidenote: The network diagramm shown at 09:00 has some slight errors. You can find a fixed version in the scene files.
