search

talafek96 / Diffusion-AD-Project

In this project we will attempt to develop a POC for detecting anomalies in images based on the ability or inability of a DDM to reconstruct them.
1 stars 1 forks source link

Clone guided-diffusion repo and manage to use it #9

Closed talafek96 closed 1 year ago

talafek96 commented 1 year ago

https://github.com/openai/guided-diffusion

Use an image of your choice (could be your cat) and find the highest noising iteration N for which it is still reconstructable. (Of course, you will have to learn how to noise the image, and reconstruct it - from different noise levels)

talafek96 commented 1 year ago

I see here that there is a function called ddimsample that is documented to sample x{t-1} from the model using DDIM. Any source/explanation for what DDIM is? Should we be interested in it or leave it alone? Might it be relevant? @EliSchwartz @aarbelle

talafek96 commented 1 year ago

@amitir22 Look at p_sample_loop_progressive, q_sample