search

YigitEkin / CLIPAway

A novel framework manipulating CLIP embeddings via projection to remove objects using Stable Diffusion prior.
https://yigitekin.github.io/CLIPAway/
MIT License
22 stars 0 forks source link

Any ETA for this #1

Open K-prog opened 3 weeks ago

K-prog commented 3 weeks ago

Hi guys, this is great work, I was going through abyildirim's work on inpainting and found this one for masked based inpainting, this seems really promising and wanted to know what would be the size of these models and when can we expect this to come out?

YigitEkin commented 3 weeks ago

Hi, thank you for your interest in our work. We plan on releasing the code and the weights of the model within a week. In terms of the model size, our best performing model is based on SD-Inpaint Pipeline which has approximately 1.1B parameters. Our proposed adapter (alpha-clip vision transformer + IP-Adapter + MLP projection block) has approximately 313M parameters which makes the total size approximately 1.4B. We also plan on releasing a gradio demo as well which in our opinion would make playing around with the model easier.