Get the latest tech news
A Deep Dive into DDPMs
deep dive into DDPMs DDPMs - Part 3 Show the code import os import scipy import numpy as np import matplotlib.pyplot as plt from PIL import Image = 1234 seed np.random.seed(seed)"ggplot")plt.style.use( In the last notebook, we discussed about Gaussian Distribution, its applications in context of diffusion models, and the forward process in diffusion models. Let’s revisit the forward process equation and the corresponding code that we saw in the last notebook.
The above equation is nice.Given the original image, we can now sample at any arbitrary timestep without simulating the entire Markov chain till that step. A neural network is sufficient to predict the mean\(\mu_{\theta}\) and the diagonal covariance matrix\(\Sigma_{\theta}\) for the reverse process as shown below in the equation, but we would also be required to frame our objectve function differently The maths above looks scary but if you look closely, we haven’t done anything fancy apart from applying standard definitions of expectation, KL-Divergence, and log to the original equation.
Or read this on Hacker News
