When I load a pre-trained model to finetune my task, even though it works well when using the model for inference for face search, it works poorly during training. I realize that this is due to the fact that the parameters (kernel) of the classification header are randomly initialized during training, so loading the pre-trained model is of limited help. So I would like to ask the authors if there are existing classification header weights?
When I load a pre-trained model to finetune my task, even though it works well when using the model for inference for face search, it works poorly during training. I realize that this is due to the fact that the parameters (kernel) of the classification header are randomly initialized during training, so loading the pre-trained model is of limited help. So I would like to ask the authors if there are existing classification header weights?