Hacker News new | ask | show | jobs
by ipunchghosts 383 days ago
Autoencoders should output these kinds of splats instead of pixel outputs and likely obtain better representations of the world at the bottleneck. These features can be used for downstream tasks.