Depth Filter vs. 3D kernels

#21
by jjjone - opened

I noticed you started incorporating depth within stable diffusion 2, which seems like a good, logical step! Have you considered just using a 3d kernel branch in the model itself? I've found 3d kernels can actually infer the depth dimension from 2d images, model depth within the kernels, and use it for better inference, especially on depth-related imaging, like hands, faces, etc. Here is the article along with code links to GitHub: https://medium.com/analytics-vidhya/fixing-stable-diffusions-achilles-heel-43da2563647e

Sign up or log in to comment