hojonathanho/diffusion
Diffusion
This project is a diffusion model training framework and image synthesis pipeline. It provides the tools necessary to train generative models to learn image data distributions through an iterative denoising process.
The framework includes a generative model evaluation tool consisting of automated scripts used to measure the quality and accuracy of produced samples.
The system covers model training pipelines and performance evaluation for generative diffusion models.
Features
- Diffusion Model Training - Implements a complete training pipeline for generative diffusion models to learn image data distributions.
- Latent Space Encoders - Utilizes compressed representations of images to reduce computational overhead during diffusion.
- Generative Model Training Tools - Provides pipelines to train generative models on image datasets using a denoising process.
- U-Net Architectures - Employs a U-Net convolutional encoder-decoder architecture with skip connections to predict and subtract noise.
- Denoising Schedulers - Provides linear and cosine variance schedules to control noise addition at each training step.
- Image Generation and Synthesis - Ships a full workflow for training and testing models that generate high-quality synthetic images.
- Iterative Noise Removal - Implements an iterative process to transform random noise into structured images by removing Gaussian noise.
- Generative Model Evaluation - Provides automated frameworks to ensure generated samples meet specific quality and performance standards.
- Model Evaluation Tools - Supplies a set of automated tools to evaluate the performance and quality of diffusion model outputs.
- Model Performance Evaluators - Includes specialized scripts to assess the quality and accuracy of generated image samples.
- Stochastic Gradient Descent - Optimizes model weights using stochastic gradient descent based on mean squared error of predicted noise.
- Diffusion Foundations - Foundational denoising diffusion probabilistic models.
- Generative Models - Denoising diffusion probabilistic models for image synthesis.
- Generation - Listed in the “Generation” section of the Awesome Diffusion Models awesome list.
