TerryMelody

> have you obtain good results finally, i also get bad results I use "clip_finetune" instead of "clip_finetune_eff", the results are better.

> Thank you for providing research materials. > > The provided source code seems to be structured so that training can proceed with the public dataset. Apart from this, if...

> @TerryMelody Unfortunatly, I have forgot this repo. However, the reason for opening this issue may same as your experience probabily. Okay thank you for your reply!

> Copy link Thank you for your suggestions! Can I understand the “image features” in your code to be the same as those implemented in Openclip source codes. It seems...

> Yes, `image features` are the `class tokens` in the original implementation. Thank you so much!