DDPM (Denoising Diffusion Probabilistic Models)

Installation Guide

Follow these steps to set up and run the DDPM project:

1. Clone the Repository git clone https://github.com/ArintonAkos/diffusion-model/

2. Install Dependencies

   • Install the required dependencies by running:

     pip install -r requirements.txt
     

   • For detailed PyTorch installation instructions, visit: PyTorch Get Started

3. Running the Code

   • Execute main.py using the command line with the desired parameters:
   • Options include:
     • -A, --all: Run setup, pre-train, and full-train steps.
     • -S, --setup: Run setup steps (visualize and data-analysis).
     • -P, --pre-train: Run pre-train steps (overfit-test and lr-find).
     • -F, --full-train: Run full-train steps (train and evaluate).
     • --model-type: Specify the model type (required). Choices: ['classification', 'ddpm'].
     • -N, --new-run: Create a new run ID.
     • --run-id: Specify a custom run ID.
     • --visualize: Visualize the dataset.
     • --data-analysis: Analyze the dataset.
     • --overfit-test: Perform an overfitting test.
     • --lr-find: Find an optimal learning rate.
     • --train: Train the model on the full dataset.
     • --evaluate: Evaluate the model on the validation set.
     • --federated: Enable federated training.
     • --model-path: Specify the path to a pre-trained model.
     • --dataset-name: Specify the dataset name.
     • --config-file: Specify the path to the configuration file. Default: 'configs/config.yml'.

Example command to train the model: python main.py --train --new-run --config=configs/homogeneous.yml --federated

Configuration

Edit the configuration file configs/config.yml to set various parameters like batch size, learning rate, etc., to suit your requirements.

Contributions

Contributions to this project are welcome. Please follow the standard Git workflow - fork, clone, feature branch, commits, and pull request.

Good results:

• Centralized train of MLP:
  • Run: ddpm_20231207-004644
  • Model: model_ep_99_20231207-010315.pth
• Federated train of MLP:
  • Run: classification-federated
  • Model: model_round_9_20231208-194853.pth

References

For more in-depth understanding and background of DDPM, refer to the following papers and resources:

• Improved Denoising Diffusion Probabilistic Models
• Denoising Diffusion Probabilistic Models
• How Diffusion Models Work - DeepLearning.AI Recap
• Conditional Diffusion MNIST GitHub Repository
• Keras Implementation of DDPM