-
Notifications
You must be signed in to change notification settings - Fork 9
/
yad2k.py
233 lines (230 loc) · 10.5 KB
/
yad2k.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
"""
Reads Darknet19 config & weights and creates Keras model with Tensorflow backend.
Currently only supports layers in Darknet19 config.
"""
# Libraries
import argparse
import configparser
import io
import os
from collections import defaultdict
import numpy as np
from keras import backend as K
from keras.layers import Conv2D, GlobalAveragePooling2D, Input, Lambda,MaxPooling2D
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.merge import concatenate
from keras.layers.normalization import BatchNormalization
from keras.models import Model
from keras.regularizers import l2
# Utilities
from yad2k.models.keras_yolo import space_to_depth_x2,space_to_depth_x2_output_shape
from config import config
def unique_config_sections(config_file):
"""
Convert all config sections to have unique names.
Adds unique suffixes to config sections for compability with configparser.
"""
section_counters = defaultdict(int)
output_stream = io.StringIO()
with open(config_file) as fin:
for line in fin:
if line.startswith('['):
section = line.strip().strip('[]')
_section = section + '_' + str(section_counters[section])
section_counters[section] += 1
line = line.replace(section, _section)
output_stream.write(line)
output_stream.seek(0)
return output_stream
def _main():
config_path = os.path.expanduser(config['model_cfg'])
weights_path = os.path.expanduser(config['model_weights'])
output_path = os.path.expanduser(config['keras_model_path'])
assert config_path.endswith('.cfg'), '{} is not a .cfg file'.format(config_path)
assert weights_path.endswith('.weights'), '{} is not a .weights file'.format(weights_path)
assert output_path.endswith('.hdf5'), 'output path {} is not a .hdf5 file'.format(output_path)
print('Config file at {}'.format(config_path))
print('Weights file at {}'.format(weights_path))
output_root = os.path.splitext(output_path)[0]
# Load weights and config.
print('Loading weights.')
weights_file = open(weights_path, 'rb')
weights_header = np.ndarray(shape=(4, ), dtype='int32', buffer=weights_file.read(16))
print('Weights Header: ', weights_header)
print('Parsing Darknet config file.')
unique_config_file = unique_config_sections(config_path)
cfg_parser = configparser.ConfigParser()
cfg_parser.read_file(unique_config_file)
print('Creating Keras model.')
image_height = int(cfg_parser['net_0']['height'])
image_width = int(cfg_parser['net_0']['width'])
# Input layer which takes input of shape (image_height, image_width, 3)
prev_layer = Input(shape=(image_height, image_width, 3))
# All layers list
all_layers = [prev_layer]
# Weight decay, default = 0.0005
weight_decay = float(cfg_parser['net_0']['decay']) if 'net_0' in cfg_parser.sections() else 5e-4
# Counter to count number of weights loaded
count = 0
# For each section in cfg_parser.sections()
for section in cfg_parser.sections():
print('Parsing section {}'.format(section))
# If section is `convolutional`
if section.startswith('convolutional'):
# Number of kernels/filters
filters = int(cfg_parser[section]['filters'])
# Kernel/filter size
size = int(cfg_parser[section]['size'])
# Stride
stride = int(cfg_parser[section]['stride'])
# Padding
pad = int(cfg_parser[section]['pad'])
# Activation
activation = cfg_parser[section]['activation']
# Batch normailzation
batch_normalize = 'batch_normalize' in cfg_parser[section]
# padding='same' is equivalent to Darknet pad=1
padding = 'same' if pad == 1 else 'valid'
# Setting weights.
# Darknet serializes convolutional weights as: [bias/beta, [gamma, mean, variance], conv_weights]
# Shape of layer previous to this convolutional layer
prev_layer_shape = K.int_shape(prev_layer)
weights_shape = (size, size, prev_layer_shape[-1], filters)
darknet_w_shape = (filters, weights_shape[2], size, size)
weights_size = np.product(weights_shape)
# If batch normailzation is enabled, print it else leave it as normal conv2d
print('conv2d', 'bn' if batch_normalize else ' ', activation, weights_shape)
conv_bias = np.ndarray(
shape=(filters, ),
dtype='float32',
buffer=weights_file.read(filters * 4))
# Increment weights counter
count += filters
# If batch normailzation is enabled
if batch_normalize:
bn_weights = np.ndarray(
shape=(3, filters),
dtype='float32',
buffer=weights_file.read(filters * 12))
# Increment weights counter
count += 3 * filters
bn_weight_list = [
bn_weights[0], # scale gamma
conv_bias, # shift beta
bn_weights[1], # running mean
bn_weights[2] # running var
]
conv_weights = np.ndarray(
shape=darknet_w_shape,
dtype='float32',
buffer=weights_file.read(weights_size * 4))
# Increment weights counter
count += weights_size
# DarkNet conv_weights are serialized Caffe-style: (out_dim, in_dim, height, width)
# We would like to set these to Tensorflow order: (height, width, in_dim, out_dim)
conv_weights = np.transpose(conv_weights, [2, 3, 1, 0])
conv_weights = [conv_weights] if batch_normalize else [ conv_weights, conv_bias ]
# Handle activation.
act_fn = None
if activation not in ['linear','leaky']:
raise ValueError('Unknown activation function `{}` in section {}'.format(activation, section))
# Create Conv2D layer
conv_layer = (Conv2D(
filters, (size, size),
strides=(stride, stride),
kernel_regularizer=l2(weight_decay),
use_bias=not batch_normalize,
weights=conv_weights,
activation=act_fn,
padding=padding))(prev_layer)
# Create batch normalization layer if enabled
if batch_normalize:
conv_layer = (BatchNormalization(weights=bn_weight_list))(conv_layer)
prev_layer = conv_layer
# If linear activation is provided
if activation == 'linear':
# Append activation layer to `all_layers`
all_layers.append(prev_layer)
# If leaky ReLU activation is provided
elif activation == 'leaky':
act_layer = LeakyReLU(alpha=0.1)(prev_layer)
prev_layer = act_layer
# Append activation layer to `all_layers`
all_layers.append(act_layer)
# If section is `maxpool`
elif section.startswith('maxpool'):
# Kernel/filter size
size = int(cfg_parser[section]['size'])
# Stride
stride = int(cfg_parser[section]['stride'])
# Append MaxPooling2D layer to `all_layers`
all_layers.append(MaxPooling2D(
padding='same',
pool_size=(size, size),
strides=(stride, stride))(prev_layer))
# Set `prev_layer` to last layer from `all_layers`
prev_layer = all_layers[-1]
# If section is `avgpool`
elif section.startswith('avgpool'):
if cfg_parser.items(section) != []:
raise ValueError('{} with params unsupported.'.format(section))
# Append GlobalAveragePooling2D layer to `all_layers`
all_layers.append(GlobalAveragePooling2D()(prev_layer))
# Set `prev_layer` to last layer from all layers
prev_layer = all_layers[-1]
# If section is `route` (Concatenation)
elif section.startswith('route'):
# IDs of all required layers
ids = [int(i) for i in cfg_parser[section]['layers'].split(',')]
# Extract all required layers
layers = [all_layers[i] for i in ids]
# If more than 1 layers exists
if len(layers) > 1:
# Concatenate all required layers
print('Concatenating route layers:', layers)
concatenate_layer = concatenate(layers)
# Append to all layers
all_layers.append(concatenate_layer)
prev_layer = concatenate_layer
# only one layer to route
else:
skip_layer = layers[0]
# Append to all layers
all_layers.append(skip_layer)
prev_layer = skip_layer
# If section is `reorg` (Lambda)
elif section.startswith('reorg'):
block_size = int(cfg_parser[section]['stride'])
assert block_size == 2, 'Only reorg with stride 2 supported.'
# Append Lambda layer to `all_layers`
all_layers.append(Lambda(
space_to_depth_x2,
output_shape=space_to_depth_x2_output_shape,
name='space_to_depth_x2')(prev_layer))
# Set `prev_layer` to last layer from `all_layers`
prev_layer = all_layers[-1]
# If section is `region`
elif section.startswith('region'):
with open('{}_anchors.txt'.format(output_root), 'w') as f:
print(cfg_parser[section]['anchors'], file=f)
# If section is net_0
elif section.startswith('net_0'):
continue
# If section is not recognized
else:
raise ValueError('Unsupported section header type: {}'.format(section))
# Create model
model = Model(inputs=all_layers[0], outputs=all_layers[-1])
# Model summary
print(model.summary())
# Save model
model.save('{}'.format(output_path))
print('Saved Keras model to {}'.format(output_path))
# Check to see if all weights have been read.
remaining_weights = len(weights_file.read()) / 4
weights_file.close()
print('Read {} of {} from Darknet weights.'.format(count, count +remaining_weights))
if remaining_weights > 0:
print('Warning: {} unused weights'.format(remaining_weights))
if __name__ == '__main__':
_main()