This is a numpy implementation of the Denoising Diffusion Probabilistic Model (DDPM), that runs at CPU with numpy or GPU with cupy (but it's better to train it on a GPU if you don't want to wait indefinitely)

• MNIST - consists of 70,000 single channel images of digits
• CIFAR-10 - consists of 60,000 different colour images in 10 classes
• UTKFace - consists of 20,000 colour images of faces

• diffusion.py - initializes and trains the model

• train_mnist.py
• train_cifar_10.py
• train_utkface.py

• simple_convnet.py - just simple convolutional model architecture
• unet.py - unet like model architecture

• activations.py - list of activation functions for the model
• losses.py - list of loss functions for the model. But used only MSE Loss
• optimizers.py - list of gradient optimizers for the model
• schedules.py - beta schedules for the DDPM
• layers - neural network layers like Dense, Conv2D, etc

• import_mnist_data.py
• import_cifar_10_data.py
• import_utkface_data.py

At the moment, implemented DDPM and DDIM samplers. The results were obtained using the DDPM sampler

• Denoising Diffusion Probabilistic Models
• Improved Denoising Diffusion Probabilistic Models
• Denoising Diffusion Implicit Models

• The Annotated Diffusion Model
• What are diffusion models?
• Denoising Diffusion Probabilistic Models (DDPM)
• How diffusion models work: the math from scratch