1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
|
#!/usr/bin/env python
'''
Caffe Python Module Testsuite for Debian Package
Copyright (C) 2016 Zhou Mo
Reference:
http://nbviewer.jupyter.org/github/BVLC/caffe/blob/master/examples/01-learning-lenet.ipynb
http://nbviewer.jupyter.org/github/BVLC/caffe/blob/master/examples/net_surgery.ipynb
'''
import numpy
import caffe
def test_lenet_benchmark():
# load lenet deploy network
model_definition = "examples/mnist/lenet.prototxt"
caffe.set_mode_cpu()
model = caffe.Net(model_definition, caffe.TEST)
# populate random data and copy it to network
batch = numpy.random.randn(64, 1, 28, 28)
model.blobs['data'].reshape(64, 1, 28, 28)
model.blobs['data'].data[...] = batch
# randomly fill parameter into network
model.params['conv1'][0].data.shape # access test, this should be (20, 1, 5, 5)
model.params['conv1'][0].flat = numpy.random.randn(20, 1, 5, 5).flat
model.params['conv1'][1].flat = numpy.random.randn(20,).flat
model.params['conv2'][0].flat = numpy.random.randn(50, 20, 5, 5).flat
model.params['conv2'][1].flat = numpy.random.randn(50,).flat
model.params['ip1'][0].flat = numpy.random.randn(500, 800).flat
model.params['ip1'][1].flat = numpy.random.randn(500,).flat
model.params['ip2'][0].flat = numpy.random.randn(10, 500).flat
model.params['ip2'][1].flat = numpy.random.randn(10,).flat
# forward this network, here we don't care neither the network parameters nor the result.
output = model.forward()
output_prob = output['prob'][0] # prob vector for the first image in batch
prediction = output_prob.argmax()
print ('This mindless benchmark ends up with an arbitary output {}'.format(prediction))
def main():
test_lenet_benchmark()
if __name__ == "__main__":
main()
|
