Open Srinivasan95D opened 3 years ago
https://github.com/AlexeyAB/darknet/issues/4133#issuecomment-544730470
I think i found an answer here. setting onlyforward=1 on [conv] sections will make sure backpropagation will not happen for that layer, Hence it acts as if the layer is frozen. is this right @AlexeyAB ?
i was wondering if there is an option to freeze specific layers during training. I have used Stopbackward but i'd liek to know if there is an option like noupdates for a certain layer.