Hello Caffe Users,
I'm training a network similar to
Siamese Network, where the input includes N image instances, and the model includes N convolution neural networks that share weights. In Siamese Network, N=2.
I have tried N=2, 3,...9, 10 for my network, and the network trains fine. I can also save the model when N<=9.
As N increases, the caffemodel file size also increases, as it should be. When N=9, the size of the caffemodel file is 2GB.
The problem is, when N=10, when trying to save the caffemodel, my ipython kernel always died.
I'm wondering if there is file size limit for the caffemodel? Or is it somehow related to GPU memory?
Thank you.