Simple Feature Computation with Caffe 1.0

The following is a concrete example of performing feature computation for a set of ten butterfly images using the Caffe 1.0 descriptor generator. This assumes that you have installed Caffe 1.0’s python bindings appropriately and have downloaded the appropriate model files as detailed in the Caffe 1.0 Default Image Net section. Once set up, the following code will compute an AlexNet descriptor:

# Import some butterfly data
# TODO: This URL is broken. Fix or find alternative example data.
urls = ["http://www.comp.leeds.ac.uk/scs6jwks/dataset/leedsbutterfly/examples/{:03d}.jpg".format(i) for i in range(1,11)]
from smqtk.representation.data_element.url_element import DataUrlElement
el = [DataUrlElement(d) for d in urls]

# Create an algorithm instance.
from smqtk_descriptors.impls.descriptor_generator.caffe1 import CaffeDescriptorGenerator
from smqtk_dataprovider.impls.data_element.file import DataFileElement
descr_generator = CaffeDescriptorGenerator(
  network_prototxt=DataFileElement("models/bvlc_reference_caffenet/deploy.prototxt"),
  network_model=DataFileElement("models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel"),
  image_mean=DataFileElement("data/ilsvrc12/imagenet_mean.binaryproto"),
)

# Compute features on the first image
result = descr_generator.generate_one_array(el[0])
print(result)
# array([ 0.        ,  0.01254855,  0.        , ...,  0.0035853 ,
#         0.        ,  0.00388408])