PyTorch SBM

This repository contains a collection of PyTorch-Lightning implementations of Score-based Generative Model presented in research papers. Contributions and suggestions of SBMs to implement are very welcomed.

Requirements

• Python >= 3.8
• PyTorch >= 1.9
• Pytorch Lightning >= 1.4.8
• CUDA enabled computing device

Installation

git clone https://github.com/vrvrv/PyTorch-SBM
cd PyTorch-SBM
pip install -r requirements.txt

How to run

python run.py experiment=ddpm_cifar10

Models

• DDPM
• NCSN
• NCSNV2
• DDIM
• Score SDE

---

DDPM

Denoising Diffusion Probabilistic Model

Authors

Jonathan Ho, Ajay Jain, Pieter Abbeel

Run Example

python run.py experiment=ddpm_cifar10

References

• w86763777's PyTorch implementation of DDPM

---

NCSN

Generative Modeling by Estimating Gradients of the Data Distribution

Authors

Yang Song, Stefano Ermon

Run Example

python run.py experiment=ncsn_cifar10

---

NCSNV2

Improved Techniques for Training Score-Based Generative Models

Authors

Yang Song, Stefano Ermon

Run Example

python run.py experiment=ncsnv2_cifar10

---

DDIM

Denoising Diffusion Implicit Model

Authors

Jiaming Song, Chenlin Meng, Stefano Ermon

Run Example

python run.py experiment=ddim_cifar10

---

Score SDE

Score-Based Generative Modeling through Stochastic Differential Equations

Authors

Yang Song, Jashcha Sohl-Dickstein, Diederik P.Kingma, Abhishek Kumar, Stefano Ermon, Ben Poole

Run Example

python run.py experiment=score_sde_cifar10

---

Comparision

We can calculate FID and IS score on each models

python test.py expeirment=ddpm_cifar10 trainer.resume_from_checkpoint=ddpm/cifar10_epoch53.ckpt

Model	FID	IS
DDPM	0	0
NCSN	0	0
NCSNV2	0	0
Score SDE	0	0