What protobuf should I use to load the trained googlenet.caffemodel from java?

[lahw...@gmail.com]

Another question about the same trained googlenet model, as a new thread for googleability:

I'm trying to run the trained googlenet model on android, and I'm throwing together a simple renderscript/filterscript implementation of the basic ops to run it with. However, I don't know enough about protobuf and caffe to extract the parameters from the data file. My understanding is that protobuf does not store what protobuf format a file is saved as; what do I use to load .caffemodel? And will the data be in C format on disk? Will it store the dimensions in the serialized blobs?

Alternately, how would I load it in another programming language, to allow conversion to a format of my choice? (I'm not happy to be using java, I just need it because it's on android.)

https://github.com/BVLC/caffe/tree/master/models/bvlc_googlenet

-- lahwran

[lahw...@gmail.com]

Answered for myself: NetParameter.

[Thomas Wood]

Hi, did you get the Renderscript implementation working on Android? Interested to know if it worked!

thanks

[lahwran]

Partially, but it was far too slow for the quick demo I was hoping for. I was counting on nets being easier to run than to train, but unfortunately it still took around a second just to run the first layer. I might have been doing something inefficiently, but I couldn't get below half a second of actual compute for the first layer, so I moved on.

-- lahwran

--
You received this message because you are subscribed to a topic in the Google Groups "Caffe Users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/caffe-users/B6J5qFZ4ah8/unsubscribe.
To unsubscribe from this group and all its topics, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/64028242-eef6-41de-9019-6f6c93405c09%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Thomas Wood]

Thank you! Interesting to know. I also found this paper: https://arxiv.org/abs/1511.07376 so I know it's possible, just difficult!