search

ali-vilab / MimicBrush

Official implementations for paper: Zero-shot Image Editing with Reference Imitation
https://xavierchen34.github.io/MimicBrush-Page/
Apache License 2.0
1.14k stars 82 forks source link

Details about training target and loss function #21

Open chrisway613 opened 2 months ago

chrisway613 commented 2 months ago

Appreciate this excellent work! But I'm confused about the training target and loss function. According to the paper, training target is to recover the masked area of source image with diffusion model, whose "responsibility" is estimating the ground truth noise. So the loss function is noise prediction combined with masked area prediction? What the loss function exactly? Can you tell me more details about it, I would be grateful!

XavierCHEN34 commented 1 month ago

The loss is for the noise prediction of the full target image.