-
Notifications
You must be signed in to change notification settings - Fork 88
/
cfg.py
249 lines (235 loc) · 10.6 KB
/
cfg.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
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
import torch
from collections import OrderedDict
def parse_cfg(cfgfile):
def erase_comment(line):
line = line.split('#')[0]
return line
blocks = []
fp = open(cfgfile, 'r')
block = None
line = fp.readline()
while line != '':
line = line.rstrip()
if line == '' or line[0] == '#':
line = fp.readline()
continue
elif line[0] == '[':
if block:
blocks.append(block)
block = OrderedDict()
block['type'] = line.lstrip('[').rstrip(']')
# set default value
if block['type'] == 'convolutional':
block['batch_normalize'] = 0
else:
line = erase_comment(line)
key,value = line.split('=')
key = key.strip()
if key == 'type':
key = '_type'
value = value.strip()
block[key] = value
line = fp.readline()
if block:
blocks.append(block)
fp.close()
return blocks
def print_cfg(blocks):
for block in blocks:
print('[%s]' % (block['type']))
for key,value in block.items():
if key != 'type':
print('%s=%s' % (key, value))
print('')
def save_cfg(blocks, cfgfile):
with open(cfgfile, 'w') as fp:
for block in blocks:
fp.write('[%s]\n' % (block['type']))
for key,value in block.items():
if key != 'type':
fp.write('%s=%s\n' % (key, value))
fp.write('\n')
def print_cfg_nicely(blocks):
print('layer filters size input output');
prev_width = 416
prev_height = 416
prev_filters = 3
out_filters =[]
out_widths =[]
out_heights =[]
ind = -2
for block in blocks:
ind = ind + 1
if block['type'] == 'net':
prev_width = int(block['width'])
prev_height = int(block['height'])
continue
elif block['type'] == 'convolutional':
filters = int(block['filters'])
kernel_size = int(block['size'])
stride = int(block['stride'])
is_pad = int(block['pad'])
pad = (kernel_size-1)/2 if is_pad else 0
width = (prev_width + 2*pad - kernel_size)/stride + 1
height = (prev_height + 2*pad - kernel_size)/stride + 1
print('%5d %-6s %4d %d x %d / %d %3d x %3d x%4d -> %3d x %3d x%4d' % (ind, 'conv', filters, kernel_size, kernel_size, stride, prev_width, prev_height, prev_filters, width, height, filters))
prev_width = width
prev_height = height
prev_filters = filters
out_widths.append(prev_width)
out_heights.append(prev_height)
out_filters.append(prev_filters)
elif block['type'] == 'maxpool':
pool_size = int(block['size'])
stride = int(block['stride'])
width = prev_width/stride
height = prev_height/stride
print('%5d %-6s %d x %d / %d %3d x %3d x%4d -> %3d x %3d x%4d' % (ind, 'max', pool_size, pool_size, stride, prev_width, prev_height, prev_filters, width, height, filters))
prev_width = width
prev_height = height
prev_filters = filters
out_widths.append(prev_width)
out_heights.append(prev_height)
out_filters.append(prev_filters)
elif block['type'] == 'avgpool':
width = 1
height = 1
print('%5d %-6s %3d x %3d x%4d -> %3d' % (ind, 'avg', prev_width, prev_height, prev_filters, prev_filters))
prev_width = 1
prev_height = 1
out_widths.append(prev_width)
out_heights.append(prev_height)
out_filters.append(prev_filters)
elif block['type'] == 'softmax':
print('%5d %-6s -> %3d' % (ind, 'softmax', prev_filters))
out_widths.append(prev_width)
out_heights.append(prev_height)
out_filters.append(prev_filters)
elif block['type'] == 'cost':
print('%5d %-6s -> %3d' % (ind, 'cost', prev_filters))
out_widths.append(prev_width)
out_heights.append(prev_height)
out_filters.append(prev_filters)
elif block['type'] == 'reorg':
stride = int(block['stride'])
filters = stride * stride * prev_filters
width = prev_width/stride
height = prev_height/stride
print('%5d %-6s / %d %3d x %3d x%4d -> %3d x %3d x%4d' % (ind, 'reorg', stride, prev_width, prev_height, prev_filters, width, height, filters))
prev_width = width
prev_height = height
prev_filters = filters
out_widths.append(prev_width)
out_heights.append(prev_height)
out_filters.append(prev_filters)
elif block['type'] == 'route':
layers = block['layers'].split(',')
layers = [int(i) if int(i) > 0 else int(i)+ind for i in layers]
if len(layers) == 1:
print('%5d %-6s %d' % (ind, 'route', layers[0]))
prev_width = out_widths[layers[0]]
prev_height = out_heights[layers[0]]
prev_filters = out_filters[layers[0]]
elif len(layers) == 2:
print('%5d %-6s %d %d' % (ind, 'route', layers[0], layers[1]))
prev_width = out_widths[layers[0]]
prev_height = out_heights[layers[0]]
assert(prev_width == out_widths[layers[1]])
assert(prev_height == out_heights[layers[1]])
prev_filters = out_filters[layers[0]] + out_filters[layers[1]]
out_widths.append(prev_width)
out_heights.append(prev_height)
out_filters.append(prev_filters)
elif block['type'] == 'region':
print('%5d %-6s' % (ind, 'detection'))
out_widths.append(prev_width)
out_heights.append(prev_height)
out_filters.append(prev_filters)
elif block['type'] == 'shortcut':
from_id = int(block['from'])
from_id = from_id if from_id > 0 else from_id+ind
print('%5d %-6s %d' % (ind, 'shortcut', from_id))
prev_width = out_widths[from_id]
prev_height = out_heights[from_id]
prev_filters = out_filters[from_id]
out_widths.append(prev_width)
out_heights.append(prev_height)
out_filters.append(prev_filters)
elif block['type'] == 'softmax':
print('%5d %-6s' % (ind, 'softmax'))
out_widths.append(prev_width)
out_heights.append(prev_height)
out_filters.append(prev_filters)
elif block['type'] == 'connected':
filters = int(block['output'])
print('%5d %-6s %d -> %3d' % (ind, 'connected', prev_filters, filters))
prev_filters = filters
out_widths.append(1)
out_heights.append(1)
out_filters.append(prev_filters)
else:
print('unknown type %s' % (block['type']))
def load_conv(buf, start, conv_model):
num_w = conv_model.weight.numel()
num_b = conv_model.bias.numel()
conv_model.bias.data.copy_(torch.from_numpy(buf[start:start+num_b])); start = start + num_b
conv_model.weight.data.copy_(torch.from_numpy(buf[start:start+num_w])); start = start + num_w
return start
def save_conv(fp, conv_model):
if conv_model.bias.is_cuda:
convert2cpu(conv_model.bias.data).numpy().tofile(fp)
convert2cpu(conv_model.weight.data).numpy().tofile(fp)
else:
conv_model.bias.data.numpy().tofile(fp)
conv_model.weight.data.numpy().tofile(fp)
def load_conv_bn(buf, start, conv_model, bn_model):
num_w = conv_model.weight.numel()
num_b = bn_model.bias.numel()
bn_model.bias.data.copy_(torch.from_numpy(buf[start:start+num_b])); start = start + num_b
bn_model.weight.data.copy_(torch.from_numpy(buf[start:start+num_b])); start = start + num_b
bn_model.running_mean.copy_(torch.from_numpy(buf[start:start+num_b])); start = start + num_b
bn_model.running_var.copy_(torch.from_numpy(buf[start:start+num_b])); start = start + num_b
conv_model.weight.data.copy_(torch.from_numpy(buf[start:start+num_w])); start = start + num_w
return start
def save_conv_bn(fp, conv_model, bn_model):
if bn_model.bias.is_cuda:
convert2cpu(bn_model.bias.data).numpy().tofile(fp)
convert2cpu(bn_model.weight.data).numpy().tofile(fp)
convert2cpu(bn_model.running_mean).numpy().tofile(fp)
convert2cpu(bn_model.running_var).numpy().tofile(fp)
convert2cpu(conv_model.weight.data).numpy().tofile(fp)
else:
bn_model.bias.data.numpy().tofile(fp)
bn_model.weight.data.numpy().tofile(fp)
bn_model.running_mean.numpy().tofile(fp)
bn_model.running_var.numpy().tofile(fp)
conv_model.weight.data.numpy().tofile(fp)
def save_conv_shrink_bn(fp, conv_model, bn_model, eps=1e-5):
if bn_model.bias.is_cuda:
bias = bn_model.bias.data - bn_model.running_mean * bn_model.weight.data / torch.sqrt(bn_model.running_var + eps)
convert2cpu(bias).numpy().tofile(fp)
s = conv_model.weight.data.size()
weight = conv_model.weight.data * (bn_model.weight.data / torch.sqrt(bn_model.running_var + eps)).view(-1,1,1,1).repeat(1, s[1], s[2], s[3])
convert2cpu(weight).numpy().tofile(fp)
else:
bias = bn_model.bias.data - bn_model.running_mean * bn_model.weight.data / torch.sqrt(bn_model.running_var + eps)
bias.numpy().tofile(fp)
s = conv_model.weight.data.size()
weight = conv_model.weight.data * (bn_model.weight.data / torch.sqrt(bn_model.running_var + eps)).view(-1,1,1,1).repeat(1, s[1], s[2], s[3])
weight.numpy().tofile(fp)
def load_fc(buf, start, fc_model):
num_w = fc_model.weight.numel()
num_b = fc_model.bias.numel()
fc_model.bias.data.copy_(torch.from_numpy(buf[start:start+num_b])); start = start + num_b
fc_model.weight.data.copy_(torch.from_numpy(buf[start:start+num_w])); start = start + num_w
return start
def save_fc(fp, fc_model):
fc_model.bias.data.numpy().tofile(fp)
fc_model.weight.data.numpy().tofile(fp)
if __name__ == '__main__':
import sys
if len(sys.argv) != 2:
print('Usage: python cfg.py model.cfg')
exit()
blocks = parse_cfg(sys.argv[1])
print_cfg_nicely(blocks)