Transfer Learning, freezing layers

[mazatov]

I'm trying to do a transfer learning for my dataset. I have only ~30 persons, but I do have a lot of images per person (~500 - 2000) . I wanted to freeze the initial layers of the model and just train the head as I'm worried about overfitting.

Would I only need to change is_training = False on this line of code?

triplet-reid/train.py

Line 286 in a538696

endpoints, body_prefix = model.endpoints(images, is_training=True)