├── README.md
└── caffe_features_classify.py


/README.md:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/jacobgil/CaffeFeaturesExample/0635c0db786ba770c0ebc37c29858f651377e0b4/README.md


--------------------------------------------------------------------------------
/caffe_features_classify.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | import sys, time, glob
 3 | #caffe_root =  "/home/vagrant/caffe/"
 4 | #sys.path.insert(0, caffe_root + 'python')
 5 | 
 6 | import caffe
 7 | from sklearn.metrics import accuracy_score
 8 | from random import shuffle
 9 | from sklearn import svm
10 | 
11 | def init_net():
12 | 	net = caffe.Classifier(caffe_root  + 'models/bvlc_reference_caffenet/deploy.prototxt',
13 | 	                       caffe_root + 'models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel')
14 | 	net.set_phase_test()
15 | 	net.set_mode_cpu()
16 | 	# input preprocessing: 'data' is the name of the input blob == net.inputs[0]
17 | 	net.set_mean('data', np.load(caffe_root + 'python/caffe/imagenet/ilsvrc_2012_mean.npy'))  # ImageNet mean
18 | 	net.set_raw_scale('data', 255)  # the reference model operates on images in [0,255] range instead of [0,1]
19 | 	net.set_channel_swap('data', (2,1,0))  # the reference model has channels in BGR order instead of RGB
20 | 	return net
21 | 
22 | def get_features(file, net):
23 | 	#print "getting features for", file
24 | 	scores = net.predict([caffe.io.load_image(file)])
25 | 	feat = net.blobs['fc8'].data[4][:,0, 0]
26 | 	return feat
27 | 
28 | def shuffle_data(features, labels):
29 | 	new_features, new_labels = [], []
30 | 	index_shuf = range(len(features))
31 | 	shuffle(index_shuf)
32 | 	for i in index_shuf:
33 | 	    new_features.append(features[i])
34 | 	    new_labels.append(labels[i])
35 | 
36 | 	return new_features, new_labels
37 | 
38 | def get_dataset(net, A_DIR, B_DIR):
39 | 	CLASS_A_IMAGES = glob.glob(A_DIR + "/*.jpg")
40 | 	CLASS_B_IMAGES = glob.glob(B_DIR + "/*.jpg")
41 | 	CLASS_A_FEATURES = map(lambda f: get_features(f, net), CLASS_A_IMAGES)
42 | 	CLASS_B_FEATURES = map(lambda f: get_features(f, net), CLASS_B_IMAGES)
43 | 	features = CLASS_A_FEATURES + CLASS_B_FEATURES
44 | 	labels = [0] * len(CLASS_A_FEATURES) + [1] * len(CLASS_B_FEATURES)
45 | 	
46 | 	return shuffle_data(features, labels)
47 | 
48 | net = init_net()
49 | x, y = get_dataset(net, sys.argv[1], sys.argv[2])
50 | l = int(len(y) * 0.4)
51 | 
52 | x_train, y_train = x[: l], y[: l]
53 | x_test, y_test = x[l : ], y[l : ]
54 | 
55 | clf = svm.SVC()
56 | clf.fit(x_train, y_train)
57 | 
58 | y_pred = clf.predict(x_test)
59 | print "Accuracy: %.3f" % accuracy_score(y_test, y_pred)
60 | 


--------------------------------------------------------------------------------