How to train diffusers for Image denoise

[abdelkareemkobo]

Hello,
I'm building a model to denoise simple object images corrupted by various noise types (Gaussian, speckle, salt-and-pepper, etc.), with only 20 clean images. The model should denoise without knowing noise type or level. My expertise is in NLP, not computer vision.
The images are grayscale images only

1. Data Augmentation:

   • Which Python libraries (e.g., imgaug, Albumentations) support adding multiple noise types for small datasets?
   • Best practices for augmenting small datasets?

2. Model Selection:

   • Are there models (e.g., U-Net, Vision Transformers) that generalize across noise types?
   • Can Stable Diffusion be used for denoising? Any guides?

3. Noise Analysis:

   • Can noise type/level be extracted using Python (e.g., OpenCV, scikit-image)?

4. Resources:

   • Any beginner-friendly tutorials or papers for denoising?

Pre-trained models or transfer learning suggestions are welcome. Thanks!