DiffusionTrainingSampleSpace
Enums · AiDotNet.Enums
What a latent diffusion model’s training sample is: an image its VAE encodes, or a latent already encoded.
For Beginners
A latent diffusion model learns to denoise a compressed “latent” version of an image, not the pixels. If you train it on images, it compresses each one first; if you already compressed them (to save time, say), tell it so and it uses them as they are.
Fields
| Field | Summary |
|---|---|
DiffusionTrainingSampleSpace Image |
The sample is an image [batch, VAE.InputChannels, height, width] (or unbatched). |
DiffusionTrainingSampleSpace Latent |
The sample is already a scaled latent [batch, LatentChannels, height / factor, width / factor] and is noised as it is. |
int value__ |