Several pre-prepared networks at Caffe

Is there an easy way (for example, without changing caffe ) to load from several pre-prepared networks into one network? A network contains several levels with the same dimensions and names as both pre-prepared networks.

I am trying to achieve this using NVidia DIGITS and Caffe.

EDIT : I thought it would be impossible to do directly from DIGITS, as confirmed by the answers. Can anyone suggest an easy way to change the DIGITS code to be able to select multiple pre-prepared networks? I checked the code a bit and thought that learning a script would be a good place to start, but I don't have in-depth knowledge of Caffe, so I'm not sure what the best / fastest way to achieve this would be.

+7
deep-learning caffe conv-neural-network pycaffe nvidia-digits
source share
2 answers

As Shai suggested, there was no way to do this, so I decided to clone the official repository and make the appropriate changes. I changed the code so that it was possible to load several pre-prepared networks using a colon as a separator.

I created a transfer request in the official repository, and my changes were then merged with the main DIGITS branch, that is, now you can use this function in DIGITS.

+6
source share

AFAIK there is no direct way to do this.
However, you can use a clean operation to load pre-prepared models and manually assign their weights to the target network. Once you have a single network with all weights initialized in accordance with various pre-prepared models, you can save it and use it as one pre-prepared model for the rest of your work.

+2
source share

All Articles