Can pretrained model use different loss function?

I want to research various loss function with the model but I don’t sure. Is it possible to apply another loss with pretrained model.

It is technically possible to apply any loss function on pretrained models.

You may need to give more information on what kinda task are you going to apply it on, what are tasks the pretaining models trained on, etc. The answers to this type of question will be subjective and may not be concrete.

1 Like

I have loaded vggface2 pretrained model for image similarity task and curious about using different loss function. The repo use euclid distance to distinguish between images but I want to use triplet loss. Is it possible?

In my understanding, Euclidean distance is a generic distance measure. It can be used internally in Triplet loss as well (For example, the distance between the anchor and positive can be calculated by Euclidean distance).

Maybe you are thinking about siamese loss function vs. triplet loss function?
If so, yes. you can apply it of course. You might be aware of FaceNet paper which applied it.

It means if I use another pretrained model such as resnet15. I can apply triplet loss as well. Thanks for your answers.