Generative Models-Hardcode

DCGAN, convolutional VAE, PixelCNN, DDPM, Diffusion with Classifier-Free guidance.

Comparison of Generative Paradigms on CIFAR-10 Section 1

DCGAN, convolutional VAE, PixelCNN, and DDPM, trained the CIFAR-10 dataset

See Notebook

Summary of Metrics for DCGAN, VAE, PixelXNN DDPM

.json summary

PixelCNN NLL Estimate During Training (Baseline)

"")

VAE NLL Estimate During Training (Baseline)

Aggregate Plots KID vs Inference Speed, Likelihood VS training Compute

Diffusion on MNIST: DDPM and Classifier-Free Guidance Section 2

Implementing and compare two diffusion-based generative models on the MNIST datase

See Notebook

Output