Skip to content

add base function for sampling #5

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
austinroose wants to merge 3 commits into
base: main
Choose a base branch
from
Open

add base function for sampling #5

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3e4e345
add base function for sampling
Jan 11, 2023
f75e7ea
imports
Jan 11, 2023
cd3671f
use numpy instead of jax to get values from normal distribution
austinroose Jan 12, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 34 additions & 0 deletions model/UNet.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -124,6 +124,40 @@ def sigma2(gamma_x):
def alpha(gamma_x):
return np.sqrt(1 - sigma2(gamma_x))

# Sample function
def sample_step(self, rng, i, T, z_t, conditioning, guidance_weight=0.):
# rng_body = jax.random.fold_in(rng, i)
# eps = jax.random.normal(rng_body, z_t.shape)
# returns Generator object that manages state and generates the random bits,
# which are then transformed into random values from useful distributions
rng = np.random.default_rng(i)
eps = rng.standard_normal(z_t.shape)
t = (T - i)/T
s = (T - i - 1) / T

g_s = self.gamma(s)
g_t = self.gamma(t)

cond = self.embedding_vectors(conditioning)

eps_hat_cond = self.score_model(
z_t,
g_t * np.ones((z_t.shape[0],), z_t.dtype),
cond,)

eps_hat_uncond = self.score_model(
z_t,
g_t * np.ones((z_t.shape[0],), z_t.dtype),
cond * 0.,)
eps_hat = (1. + guidance_weight) * eps_hat_cond - guidance_weight * eps_hat_uncond


a = nn.sigmoid(g_s)
b = nn.sigmoid(g_t)
c = -np.expm1(g_t - g_s)
sigma_t = np.sqrt(sigma2(g_t))
z_s = np.sqrt(a / b) * (z_t - sigma_t * c * eps_hat) + np.sqrt((1. - a) * c) * eps
return z_s

class ResNet(nn.Module):
def __init__(self, in_ch, out_ch, num_blocks=4, num_layers=4, num_filters=64, kernel_size=3, stride=1, padding=1,
Expand Down