How to get probabilities for classifier (custom trained model)

[Alex Potérya]

Hello,

I've trained a model with only one custom keywords on ~ 5000 images. Now, I'm running classifier.py like this:

python python/classfier.py --pretrained_model=./models/bvlc_googlenet/bvlc_googlenet_my_model.caffemodel --model_def=./models/bvlc_googlenet/deploy.prototxt --mean_file=./data/ilsvrc12/imagenet_mean.nyp ./test.jpg ./output.txt

Obviously, it classifies it as "0" (my only keyword). But where do I get probability for this classification (same as it's done in web demo)?

Thanks!

[Alex Potérya]

Any advice?

Thanks.

[Alex Potérya]

I will extend the question to the Flickr Style fine tune model.

Is there an example showing how to work with output.txt.npy to get a list of keywords/probabilities for Flickr Style data set?

On Saturday, April 11, 2015 at 6:09:30 PM UTC-4, Alex Potérya wrote:

[CaffeStudent]

I think it's worth taking a look at this ipython notebook!

http://nbviewer.ipython.org/github/bvlc/caffe/blob/master/examples/classification.ipynb

[Alex Potérya]

Thanks! I've read this document multiple times. It doesn't seem to cover probabilities topic at all.

[CaffeStudent]

The plots they are showing are of the probability outputs, I believe.  If you pick the max from that plot, that is the probability of that class.