spreck / stable-diffusion-regularization-images

Stable Diffusion Regularization Images in 512px, 768px and 1024px on 1.5, 2.1 and SDXL 1.0 checkpoints

Stable Diffusion Regularization Image Dataset

Pre-rendered regularization images of man and women on Stable Diffusion 1.5, 2.1 and SDXL checkpoints

What are Regularization Images?

Regularization images are images that are used as part of a regularization process to improve the stability and performance of deep learning models.

By creating regularization images, you're essentially defining a "class" of what you're trying to invert. For example, if you're trying to invert a new airplane, you might want to create a bunch of airplane images for regularization. This is so that your training doesn't drift into another class, let's say "car" or "bike". This can even help against going towards a "toy plane" if you are using real references and not interpretations. source

About the images:

All images were generated with Stable Diffusion 1.5, 2.1, and SDXL 1.0 checkpoint. For each class (man and woman), a total of 5000 images were generated.

Class of images:

• woman
• man

How the images are generated:

All images were generated using only the base Stable Diffusion checkpoints (1.5, 2.1 and SDXL), with simple prompts such as photo of a woman, a woman or simply woman.

Other related projects:

• https://github.com/tobecwb/stable-diffusion-face-dataset
• https://github.com/tobecwb/facial-features-yolo8x-seg
• https://github.com/tobecwb/stable-diffusion-regularization-images