search

vision-x-nyu / image-sculpting

Code release for Image Sculpting: Precise Object Editing with 3D Geometry Control [CVPR 2024]
https://image-sculpting.github.io/
MIT License
268 stars 19 forks source link

Training data for image enhancement #8

Open YiruS opened 1 month ago

YiruS commented 1 month ago

Hi there,

First of all, thanks so much for providing the high-quality results and detailed codebase!

There's one question that seems not mention in the paper. What is the training data that is used for DDIM plus process (image enhancement process)? And how many images were used in the fine-tuning?

Thanks and looking forward to your reply!

domejiraphon commented 1 month ago

Hi, we are fine-tuning the model using DreamBooth. The code we used is from Diffusers implementation For the instance image, we only use the input image and generate a few hundred images for the prior loss.