asuradayuci / tf-clip Goto Github PK

在CLIP-Memory module中的前几步几乎与VIT一样啊，为什么不直接写使用VIT呢？

Hi！Thank you for the amazing work!
I have a question about the paper. You propose a noval Text-free framework based on CLIP, removing the original CLIP Text-Encoder. However, one of the major advantages of CLIP Text-Encoder is the open-vocabulary ability.
With your method using vision memory, the recogntion is limited to the features of Y identities. If the model encounters a probe for a new identity during testing, how does it handle it? I understand that this method actually only utilizes the matching concept of CLIP and does not benefit from CLIP's open-vocabulary abiltiy?

可否添加一个文件夹结构树，包含数据集、模型权重存放位置等，代码中包含作者的文件夹目录，但我并不明白文件夹其中之间的关系，非常感谢

Hello! Thank you for sharing this interesting repository. I would like to try out this project, is there any pretrained models available for download?