@LiDaiY if you initialized the parameters of the resnet with the flag pretrained=True then the model would already have learnt a representation. You might get better results, but the purpose of this repository is to learn a good representation from scratch using BYOL.
The other model parameters (classifier, mlp head, ...) are initialized randomly (there is not really any other option).
@LiDaiY if you initialized the parameters of the resnet with the flag
pretrained=Truethen the model would already have learnt a representation. You might get better results, but the purpose of this repository is to learn a good representation from scratch using BYOL. The other model parameters (classifier, mlp head, ...) are initialized randomly (there is not really any other option).