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) and standard score
matching path samples.
Module Contents
-
class gensbi.diffusion.path.path_sample.EDMPathSample[source]
Represents a sample of a diffusion generated probability path.
-
x_1[source]
the target sample \(X_1\).
- Type:
Array
-
sigma[source]
the noise scale \(t\).
- Type:
Array
-
x_t[source]
samples \(X_t \sim p_t(X_t)\), shape (batch_size, …).
- Type:
Array
-
get_batch()[source]
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[source]
-
x_1: jax.Array[source]
-
x_t: jax.Array[source]
-
class gensbi.diffusion.path.path_sample.SMPathSample[source]
Represents a sample from a standard score matching probability path.
The noising process is: \(x_t = \mu(t) x_1 + \sigma(t) \epsilon\)
-
x_1[source]
the clean target sample, shape (batch_size, …).
- Type:
Array
-
x_t[source]
the noised sample, shape (batch_size, …).
- Type:
Array
-
t[source]
the diffusion time, shape (batch_size, …).
- Type:
Array
-
noise[source]
the Gaussian noise epsilon used for noising, shape (batch_size, …).
- Type:
Array
-
std_t[source]
the marginal standard deviation at time t, shape (batch_size, …).
- Type:
Array
-
get_batch()[source]
Returns the batch as a tuple (x_1, x_t, t, noise, std_t).
- Returns:
The clean sample, noised sample, time, noise, and marginal std.
- Return type:
Tuple[Array, Array, Array, Array, Array]
-
noise: jax.Array[source]
-
std_t: jax.Array[source]
-
t: jax.Array[source]
-
x_1: jax.Array[source]
-
x_t: jax.Array[source]
