gensbi.diffusion.path.path_sample

Path sample data structures for diffusion models.

This module defines data structures for representing samples along the diffusion probability path, including EDM path samples from “Elucidating the Design Space of Diffusion-Based Generative Models” (Karras et al., 2022).

Classes

EDMPathSample

Represents a sample of a diffusion generated probability path.

Module Contents

class gensbi.diffusion.path.path_sample.EDMPathSample

Represents a sample of a diffusion generated probability path.

x_1

the target sample \(X_1\).

Type:

Array

sigma

the noise scale \(t\).

Type:

Array

x_t

samples \(X_t \sim p_t(X_t)\), shape (batch_size, …).

Type:

Array

get_batch()

Returns the batch as a tuple (x_1, x_t, sigma).

Returns:

Tuple[Array, Array, Array]

Return type:

The target sample, the noisy sample, and the noise scale.

sigma: jax.Array

x_1: jax.Array

x_t: jax.Array