.caffemodel and .solverstate sizes

[Ákos Godó]

Hello everyone!

I've got a few a questions to ask about the .caffemodel and .solverstate files that pop out of my training process, namely:

• What exactly is contained within these files?
• How can I open/parse them to see for myself?
• Why in the world are these so HUGE(for the ImageNet training they are about 240MB)?

Thank you in advance and thank you for the great work on the caffe project!

[sharath s]

Mr. Akos Godo,

Did you find answer for this?

I am stuck in the same situtation now, I would really appreciate if you did the above three points you mentioned.

Thanks 

[Ákos Godó]

Dear Mr Sharath S,

It would appear that these files contain the current weights for the network. You can in fact visualize the contents of these files using the filter visualization python example.
The size makes sense if you take each weight as a float.

I hope this helped,
Ákos

[sharath s]

Akos Godo,

Where can i find this "filter visualization python example" in caffe?

do you also know how can we map the trained model on a DET Curve?

Thanks

BR
Sharath

[Ákos Godó]

Dear Sharath,

You can find it here:
http://nbviewer.ipython.org/github/BVLC/caffe/blob/master/examples/00-classification.ipynb

You can find other scripts in your /examples directory which is in the caffe directory.

You can also use the output of this script to help you plot a DET curve with a little tinkering.

Hope that helped,
Ákos

[sharath s]

Thanks a lot Mr. Akos Godos,

I will try that.