I have tried many iterations but still cannot get satisfactory total loss, any idea where is stuck? I have checked checkpoint restore many times, and they seem fine. I’m not so confident on the variables to fine tune, in the assignment, it hinted 4 variables.
Please click my name and message your notebook as an attachment along with the tensorflow version in use.
pip list | grep tensorflow
In this assignment you have to create a model checkpoint composed of the feature extractor and the bounding box prediction layer, since you want to detect a new class that is not present in the pretrained dataset. After you restore this checkpoint you choose to fine tune the last 2 layers (which include the layers weights and biases) for the bounding box regression prediction and the classification. The special care should be paid in creating the model checkpoint and you need to go through those steps carefully to choose the right layers, all else should fall into place if that part of the process is done correctly.