gaopengcuhk/Tip-Adapter

Where is the CLIP-Adapter code, please?

Closed this issue · 5 comments

Greetings! Really fantastic work!
But where is the CLIP-Adapter code? I wonder how is the two-layer MLP in CLIP-Adapter structured.

According to your papers, there is a two-layer MLP adapter for the language branch in CLIP-Adapter, but there is not in TIP-Adapter. Am I right?

Thanks for your interest on our work. The CLIP-Adapter code is partly released at here, where shows how two-layer MLP structured. We will complete it soon.

https://github.com/gaopengcuhk/CLIP-Adapter

Yes, but the best-performing version of CLIP-Adapter only contains visual adapter like Tip-Adapter.

I see. It seems there is also no adapter for the text branch in CLIP-Adapter.
Is this because the text adapter is useless compared with the image adapter?

Yeah I found corresponding experimental results in the CLIP-Adapter paper. Thanks anyway!