So, I have a GoogleNet model that appears to be missing it’s main FC layer. I need the FC layer for calculations, and thus I need to find a way to recreate the missing layer.
I have a slightly modified version of the dataset (images were resized to be a square) that the model was trained on, but it’s 18 million images in total. This means that retraining a new FC layer while freezing the rest of the model’s parameters will take a long time, and thus I was wondering if there are any faster ways that I could go about this?