reppy4620/diffusion

My implementation of diffusion (like) models

diffusion

The Implementation of DIffusion (like) models for fun.

• DDPM(https://arxiv.org/abs/2006.11239)
• Score-SDE(https://arxiv.org/abs/2011.13456)
• Rectified Flow(https://arxiv.org/abs/2209.03003)
• Conditional Flow Matching(Simplified version)(https://arxiv.org/abs/2302.00482)
• Consistency Models(https://arxiv.org/abs/2303.01469)

Requirements

pip install torch torchvision einops numpy datasets tqdm scipy omegaconf

Result

• All the datasets are loaded with huggingface/datasets.

• MNIST

  • 20 epoch
  • Unet with 1 up-down block and mid block

• FASHION MNIST

  • 20 epoch
  • Unet with 1 up-down block and mid block

• CIFAR10

  • 100 epoch
  • Unet with 2 up-down block and mid block

• AFHQ - CAT

  • AFHQ dataset filtered with cat label
  • 300 epoch
  • Unet with 2 up-down block and mid block

Model	MNIST	FASHION MNIST	CIFAR10	CAT
DDPM
Score-SDE
Rectified Flow

• Score-SDE is not stable with regards to color tones in the CAT dataset.
  • Since the previous experiment worked well at times, it may depend on the model size.
  • This phenomenon sometimes occurs in DDPM.