Finding Accuracy of CaffeModel

[John Smith]

Hello everyone,

I'm quite new to caffe but I've reached the point of playing around with it and successfully train a model.

I trained a model for up to 100k iterations and had it saved at every 10k iterations as a caffemodel.

Now my question is how do I know out of the 10 different 10k iteration caffemodel files has the highest accuracy? 

I know in the terminal it outputs this and actually noticed the accuracy going down as the iterations got higher (overfitting). Is there a way to view the caffemodel files and see what it's accuracy is? If not, an option in caffe to save the accuracy to the filename? 

Thanks,

John

[Przemek D]

Accuracy is not stored within the caffemodel. You could store the output log to a file using standard linux redirect mechanism (eg. caffe train --solver "solver.prototxt" > output.log) and correlate the snapshot saving messages with test accuracy outputs. Right now your only option seems to be to load each snapshot individually and test it.