about training code & details.

[Weifeng]

I'm trying to replicate your result. Is there any plan to release the training code? I have to doubts about your work: Q1: which part of id_encoder got trained based on image_encoder? Q2: is the lora same as the one in ip-adapter?

[Zhen Li]

Thanks for your interest! We are in the process of organizing the training code, but we cannot guarantee the release time.
Q1: CLIP image encoder
Q2: The LoRA size in ip-adapter-FaceID is much bigger than ours.

[Weifeng]

Thanks for your interest! We are in the process of organizing the training code, but we cannot guarantee the release time. Q1: CLIP image encoder Q2: The LoRA size in ip-adapter-FaceID is much bigger than ours.

Thanks! Looking forward to the training code. So, you train the whole CLIP image encoder, unlike part layers in paper?