initialized by other weight

[wangyf8848]

Hello, may I ask if you think the linear layer used for the network middle layer is to train the text-language alignment ability from scratch? And have you tried to train the framwork with initialized by other fine pretrained feature extractor（have not been trained by clip）？

[ByChelsea]

In fact, I cannot assert with certainty the behavior of the intermediate layers, but what can be confirmed is that fine-tuning is indeed very useful. Therefore, you can also try other feature extractors, such as DINO.

[wangyf8848]

Ok, thank you very much for your reply!