The pre-training datasets?

[zenithc-git]

Hello,

I am currently studying the code of your project and I am wondering what the pre-training dataset used for the model is. Could you please provide more details about the pre-training dataset?
It would be very helpful for my understanding.

Thank you!

[azshue]

Hi @zenithc-git ,

Thank you for your interest in our work. TPT is a test-time method, so we didn't do any model training. In the paper, we mainly use OpenAI's CLIP model as our pre-trained VLM. Please refer to the original work (or open-source alternatives like https://github.com/mlfoundations/open_clip) for more details about the pre-training dataset. Hope this helps. Thanks!