DiffusionModel (WIP)

• Re-implementing Denoising Diffusion Probabilistic Models and Denoising Diffusion Implicit Models using Pytorch

Concept of DDPM & DDIM

• DDPM

<p align="center"> <img width="800" alt="concept_of_ddpm" src="https://user-images.githubusercontent.com/22078438/178132101-035afa58-d37b-43ef-ae04-0616a8c044d1.png"> </p>

• DDIM

<p align="center"> <img width="800" alt="concept_of_ddim" src="https://user-images.githubusercontent.com/22078438/178132102-202ba6f4-febf-4b52-881c-73d391546f75.png"> </p>

Model Architecture (WIP)

Trining Process (WIP)

Evaluation Process (WIP)

Reference

• Denoising Diffusion Probabilistic Models (Paper link)
• Denoising Diffusion Implicit Models (Paper link)

Usage of Training

• DDPM

python main.py --model_type=ddpm

• DDIM

python main.py --model_type=ddim

Get a Awesome Generating Images ! (WIP)

• DDPM
• DDIM

Experiments

• Quantitative result

| Model | FID | IS | #Params | | :---: | :---: | :---: | :---: | | DDPM | - | - | - | | DDIM | - | - | - |

• Qualitative result (WIP, attach more images later !, Below images are trained model result !)

  • DDPM

  

  • DDIM

  

Interpolations Experiments (WIP)

• Example (Below image is paper result)

<img width="828" alt="interpolation_imgs" src="https://user-images.githubusercontent.com/22078438/178132052-9fb621b8-dda6-4298-afaa-ce0437c6cecd.png">

Metric (FID) Evaluation

• Usage

python metric/fid_test.py --cuda=True

Metric (IS) Evalution (WIP)

• Usage

WIP

---

@article{ho2020denoising,
  title={Denoising diffusion probabilistic models},
  author={Ho, Jonathan and Jain, Ajay and Abbeel, Pieter},
  journal={Advances in Neural Information Processing Systems},
  volume={33},
  pages={6840--6851},
  year={2020}
}

@article{song2020denoising,
  title={Denoising diffusion implicit models},
  author={Song, Jiaming and Meng, Chenlin and Ermon, Stefano},
  journal={arXiv preprint arXiv:2010.02502},
  year={2020}
}