Generative_Modeling_Project

Report containing the code and the report of the projet 5 PSEUDOINVERSE-GUIDED DIFFUSION MODELS FOR INVERSE PROBLEMS of Generative Models for image @ MVA (B. GALERNE & A. LECLAIRE)

Structure

Generative_modeling_project/
├── README.md
├── init.ipynb
├── setup.py
├── data/
│   └── __init__.py
├── figs/
│   └── __init__.py
├── ntbk/
│   ├── __init__.py
│   ├── evaluator.ipynb
│   └── tp_6.ipynb
├── src/
│   ├── __init__.py
│   ├── ddpm.py
│   ├── eval.py
│   ├── h_fcn.py
│   ├── h_utils.py
│   ├── pigdm.py
│   └── utils.py
└── tests/
    ├── __init__.py
    └── test.ipynb

• src: contains all the function of the project
• figs: contains some outputs of the model
• tests: sandbox with jupyter notebooks
• ntbk: notebooks to present our main results and how to reproduce them
• data: containing all the data and the pre trained models

Get Start

First, please run:

chmod +x init_repo.sh

Then

./init_repo.sh

from a terminal. Or simply run the whole notebook init.ipynb to download the packages, the pre-trained models and the data

Experiments

To reproduce the experiments, you can either start the files in ntbk/ and run them or execute the following commands

cd scripts

Config the parameters on the file config.yaml then run:

python evaluator.py

/!\ Important: In the notebooks and script, we refer to $\Pi GDM$ as "pigdm" and $DPS$ as "ddim", even though ddim is the name of the sampler, we consider that dps is the baseline so every variable named as ddim refer to dps !!

The official implementation of $\Pi$GDM can be found here