forked from wwwanghao/caffe2pytorch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
verify_deploy.py
147 lines (133 loc) · 6.35 KB
/
verify_deploy.py
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
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
# 2017.12.16 by xiaohang
import sys
from caffenet import *
import numpy as np
import argparse
import torch.nn as nn
from torch.autograd import Variable
from torch.nn.parameter import Parameter
import time
def load_image(imgfile):
image = caffe.io.load_image(imgfile)
transformer = caffe.io.Transformer({'data': (1, 3, args.height, args.width)})
transformer.set_transpose('data', (2, 0, 1))
transformer.set_mean('data', np.array([args.meanB, args.meanG, args.meanR]))
transformer.set_raw_scale('data', args.scale)
transformer.set_channel_swap('data', (2, 1, 0))
image = transformer.preprocess('data', image)
image = image.reshape(1, 3, args.height, args.width)
return image
def load_synset_words(synset_file):
lines = open(synset_file).readlines()
synset_dict = dict()
for i, line in enumerate(lines):
synset_dict[i] = line.strip()
return synset_dict
def forward_pytorch(protofile, weightfile, image):
net = CaffeNet(protofile, width=args.width, height=args.height, omit_data_layer=True, phase='TEST')
if args.cuda:
net.cuda()
print(net)
net.load_weights(weightfile)
net.eval()
image = torch.from_numpy(image)
if args.cuda:
image = Variable(image.cuda())
else:
image = Variable(image)
t0 = time.time()
blobs = net(image)
t1 = time.time()
return t1-t0, blobs, net.models
# Reference from:
def forward_caffe(protofile, weightfile, image):
if args.cuda:
caffe.set_device(0)
caffe.set_mode_gpu()
else:
caffe.set_mode_cpu()
net = caffe.Net(protofile, weightfile, caffe.TEST)
net.blobs['data'].reshape(1, 3, args.height, args.width)
net.blobs['data'].data[...] = image
t0 = time.time()
output = net.forward()
t1 = time.time()
return t1-t0, net.blobs, net.params
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='convert caffe to pytorch')
parser.add_argument('--protofile', default='', type=str)
parser.add_argument('--weightfile', default='', type=str)
parser.add_argument('--imgfile', default='', type=str)
parser.add_argument('--height', default=224, type=int)
parser.add_argument('--width', default=224, type=int)
parser.add_argument('--meanB', default=104, type=float)
parser.add_argument('--meanG', default=117, type=float)
parser.add_argument('--meanR', default=123, type=float)
parser.add_argument('--scale', default=255, type=float)
parser.add_argument('--synset_words', default='', type=str)
parser.add_argument('--cuda', action='store_true', help='enables cuda')
args = parser.parse_args()
print(args)
protofile = args.protofile
weightfile = args.weightfile
imgfile = args.imgfile
image = load_image(imgfile)
time_pytorch, pytorch_blobs, pytorch_models = forward_pytorch(protofile, weightfile, image)
time_caffe, caffe_blobs, caffe_params = forward_caffe(protofile, weightfile, image)
print('pytorch forward time %d', time_pytorch)
print('caffe forward time %d', time_caffe)
layer_names = pytorch_models.keys()
blob_names = pytorch_blobs.keys()
print('------------ Parameter Difference ------------')
for layer_name in layer_names:
if type(pytorch_models[layer_name]) in [nn.Conv2d, nn.Linear, Scale, Normalize]:
pytorch_weight = pytorch_models[layer_name].weight.data
if args.cuda:
pytorch_weight = pytorch_weight.cpu().numpy()
else:
pytorch_weight = pytorch_weight.numpy()
caffe_weight = caffe_params[layer_name][0].data
weight_diff = abs(pytorch_weight - caffe_weight).sum()
if type(pytorch_models[layer_name].bias) == Parameter:
pytorch_bias = pytorch_models[layer_name].bias.data
if args.cuda:
pytorch_bias = pytorch_bias.cpu().numpy()
else:
pytorch_bias = pytorch_bias.numpy()
caffe_bias = caffe_params[layer_name][1].data
bias_diff = abs(pytorch_bias - caffe_bias).sum()
print('%-30s weight_diff: %f bias_diff: %f' % (layer_name, weight_diff, bias_diff))
else:
print('%-30s weight_diff: %f' % (layer_name, weight_diff))
elif type(pytorch_models[layer_name]) == nn.BatchNorm2d:
if args.cuda:
pytorch_running_mean = pytorch_models[layer_name].running_mean.cpu().numpy()
pytorch_running_var = pytorch_models[layer_name].running_var.cpu().numpy()
else:
pytorch_running_mean = pytorch_models[layer_name].running_mean.numpy()
pytorch_running_var = pytorch_models[layer_name].running_var.numpy()
caffe_running_mean = caffe_params[layer_name][0].data/caffe_params[layer_name][2].data[0]
caffe_running_var = caffe_params[layer_name][1].data/caffe_params[layer_name][2].data[0]
running_mean_diff = abs(pytorch_running_mean - caffe_running_mean).sum()
running_var_diff = abs(pytorch_running_var - caffe_running_var).sum()
print('%-30s running_mean_diff: %f running_var_diff: %f' % (layer_name, running_mean_diff, running_var_diff))
print('------------ Output Difference ------------')
for blob_name in blob_names:
if args.cuda:
pytorch_data = pytorch_blobs[blob_name].data.cpu().numpy()
else:
pytorch_data = pytorch_blobs[blob_name].data.numpy()
caffe_data = caffe_blobs[blob_name].data
diff = abs(pytorch_data - caffe_data).sum()
print('%-30s pytorch_shape: %-20s caffe_shape: %-20s output_diff: %f' % (blob_name, pytorch_data.shape, caffe_data.shape, diff/pytorch_data.size))
if args.synset_words != '':
print('------------ Classification ------------')
synset_dict = load_synset_words(args.synset_words)
if 'prob' in blob_names:
if args.cuda:
pytorch_prob = pytorch_blobs['prob'].data.cpu().view(-1).numpy()
else:
pytorch_prob = pytorch_blobs['prob'].data.view(-1).numpy()
caffe_prob = caffe_blobs['prob'].data[0]
print('pytorch classification top1: %f %s' % (pytorch_prob.max(), synset_dict[pytorch_prob.argmax()]))
print('caffe classification top1: %f %s' % (caffe_prob.max(), synset_dict[caffe_prob.argmax()]))