- This is an simple implementation of a Diffusion model (Stable diffusion) in Tensorflow & Keras.
- This code is used in Kaggle
- I'm using emoji images for each company such as Apple, Facebook, Twitter etc.
- I only used the 400 first images (emojis of person/faces etc.. no flags of symbols)
- Link to download : https://www.kaggle.com/datasets/subinium/emojiimage-dataset
- The model is built on a UNET architecture
- We first generate a random image noisy at a certain time step, along with the accompanying noise.
- We then feed the noisy image to the Unet model to try to denoise it, and calculate the error with the accompanying noise.