def _main(args):
config_path = os.path.expanduser(args.config_path)
weights_path = os.path.expanduser(args.weights_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)
output_path = os.path.expanduser(args.output_path)
assert output_path.endswith(
'.h5'), 'output path {} is not a .h5 file'.format(output_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)
# TODO: Check transpose flag when implementing fully connected layers.
# transpose = (weight_header[0] > 1000) or (weight_header[1] > 1000)
print('Parsing Darknet config.')
unique_config_file = unique_config_sections(config_path)
cfg_parser = configparser.ConfigParser()
cfg_parser.read_file(unique_config_file)
print('Creating Keras model.')
if args.fully_convolutional:
image_height, image_width = None, None
else:
image_height = int(cfg_parser['net_0']['height'])
image_width = int(cfg_parser['net_0']['width'])
prev_layer = Input(shape=(image_height, image_width, 3))
all_layers = [prev_layer]
weight_decay = float(cfg_parser['net_0']['decay']
) if 'net_0' in cfg_parser.sections() else 5e-4
count = 0
for section in cfg_parser.sections():
print('Parsing section {}'.format(section))
if section.startswith('convolutional'):
filters = int(cfg_parser[section]['filters'])
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
pad = int(cfg_parser[section]['pad'])
activation = cfg_parser[section]['activation']
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]
prev_layer_shape = K.int_shape(prev_layer)
# TODO: This assumes channel last dim_ordering.
weights_shape = (size, size, prev_layer_shape[-1], filters)
darknet_w_shape = (filters, weights_shape[2], size, size)
weights_size = np.product(weights_shape)
print('conv2d', 'bn'
if batch_normalize else ' ', activation, weights_shape)
conv_bias = np.ndarray(
shape=(filters, ),
dtype='float32',
buffer=weights_file.read(filters * 4))
count += filters
if batch_normalize:
bn_weights = np.ndarray(
shape=(3, filters),
dtype='float32',
buffer=weights_file.read(filters * 12))
count += 3 * filters
# TODO: Keras BatchNormalization mistakenly refers to var
# as std.
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))
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)
# TODO: Add check for Theano dim ordering.
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 == 'leaky':
pass # Add advanced activation later.
elif activation != 'linear':
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)
if batch_normalize:
conv_layer = (BatchNormalization(
weights=bn_weight_list))(conv_layer)
prev_layer = conv_layer
if activation == 'linear':
all_layers.append(prev_layer)
elif activation == 'leaky':
act_layer = LeakyReLU(alpha=0.1)(prev_layer)
prev_layer = act_layer
all_layers.append(act_layer)
elif section.startswith('maxpool'):
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
all_layers.append(
MaxPooling2D(
padding='same',
pool_size=(size, size),
strides=(stride, stride))(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('avgpool'):
if cfg_parser.items(section) != []:
raise ValueError('{} with params unsupported.'.format(section))
all_layers.append(GlobalAveragePooling2D()(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('route'):
ids = [int(i) for i in cfg_parser[section]['layers'].split(',')]
layers = [all_layers[i] for i in ids]
if len(layers) > 1:
print('Concatenating route layers:', layers)
concatenate_layer = concatenate(layers)
all_layers.append(concatenate_layer)
prev_layer = concatenate_layer
else:
skip_layer = layers[0] # only one layer to route
all_layers.append(skip_layer)
prev_layer = skip_layer
elif section.startswith('reorg'):
block_size = int(cfg_parser[section]['stride'])
assert block_size == 2, 'Only reorg with stride 2 supported.'
all_layers.append(
Lambda(
space_to_depth_x2,
output_shape=space_to_depth_x2_output_shape,
name='space_to_depth_x2')(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('region'):
with open('{}_anchors.txt'.format(output_root), 'w') as f:
print(cfg_parser[section]['anchors'], file=f)
elif (section.startswith('net') or section.startswith('cost') or
section.startswith('softmax')):
pass # Configs not currently handled during model definition.
else:
raise ValueError(
'Unsupported section header type: {}'.format(section))
# Create and save model.
model = Model(inputs=all_layers[0], outputs=all_layers[-1])
print(model.summary())
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 args.plot_model:
plot(model, to_file='{}.png'.format(output_root), show_shapes=True)
print('Saved model plot to {}.png'.format(output_root))
def _main(args):
config_path = os.path.expanduser(args.config_path)
weights_path = os.path.expanduser(args.weights_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)
output_path = os.path.expanduser(args.output_path)
assert output_path.endswith(
'.h5'), 'output path {} is not a .h5 file'.format(output_path)
output_root = os.path.splitext(output_path)[0]
# Load weights and config.
print('Loading weights.')
weights_file = open(weights_path, 'rb')
major, minor, revision = np.ndarray(shape=(3, ),
dtype='int32',
buffer=weights_file.read(12))
if (major * 10 + minor) >= 2 and major < 1000 and minor < 1000:
seen = np.ndarray(shape=(1, ),
dtype='int64',
buffer=weights_file.read(8))
else:
seen = np.ndarray(shape=(1, ),
dtype='int32',
buffer=weights_file.read(4))
print('Weights Header: ', major, minor, revision, seen)
print('Parsing Darknet config.')
unique_config_file = unique_config_sections(config_path)
cfg_parser = configparser.ConfigParser()
cfg_parser.read_file(unique_config_file)
print('Creating Keras model.')
input_layer = Input(shape=(None, None, 3))
prev_layer = input_layer
all_layers = []
weight_decay = float(cfg_parser['net_0']['decay']
) if 'net_0' in cfg_parser.sections() else 5e-4
count = 0
out_index = []
for section in cfg_parser.sections():
print('Parsing section {}'.format(section))
if section.startswith('convolutional'):
filters = int(cfg_parser[section]['filters'])
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
pad = int(cfg_parser[section]['pad'])
activation = cfg_parser[section]['activation']
batch_normalize = 'batch_normalize' in cfg_parser[section]
padding = 'same' if pad == 1 and stride == 1 else 'valid'
# Setting weights.
# Darknet serializes convolutional weights as:
# [bias/beta, [gamma, mean, variance], conv_weights]
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)
print('conv2d', 'bn' if batch_normalize else ' ', activation,
weights_shape)
conv_bias = np.ndarray(shape=(filters, ),
dtype='float32',
buffer=weights_file.read(filters * 4))
count += filters
if batch_normalize:
bn_weights = np.ndarray(shape=(3, filters),
dtype='float32',
buffer=weights_file.read(filters * 12))
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))
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])
## use new classes
if filters == 255:
filters = 24
conv_weights, conv_bias, bn_weight_list, bn_weights = compare_new_classnum(
conv_weights, conv_bias, bn_weight_list, bn_weights)
conv_weights = [conv_weights] if batch_normalize else [
conv_weights, conv_bias
]
# Handle activation.
act_fn = None
if activation == 'leaky':
pass # Add advanced activation later.
elif activation != 'linear':
raise ValueError(
'Unknown activation function `{}` in section {}'.format(
activation, section))
# Create Conv2D layer
if stride > 1:
# Darknet uses left and top padding instead of 'same' mode
prev_layer = ZeroPadding2D(((1, 0), (1, 0)))(prev_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)
if batch_normalize:
conv_layer = (BatchNormalization(
weights=bn_weight_list))(conv_layer)
prev_layer = conv_layer
if activation == 'linear':
all_layers.append(prev_layer)
elif activation == 'leaky':
act_layer = LeakyReLU(alpha=0.1)(prev_layer)
prev_layer = act_layer
all_layers.append(act_layer)
elif section.startswith('route'):
ids = [int(i) for i in cfg_parser[section]['layers'].split(',')]
layers = [all_layers[i] for i in ids]
if len(layers) > 1:
print('Concatenating route layers:', layers)
concatenate_layer = Concatenate()(layers)
all_layers.append(concatenate_layer)
prev_layer = concatenate_layer
else:
skip_layer = layers[0] # only one layer to route
all_layers.append(skip_layer)
prev_layer = skip_layer
elif section.startswith('maxpool'):
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
all_layers.append(
MaxPooling2D(pool_size=(size, size),
strides=(stride, stride),
padding='same')(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('shortcut'):
index = int(cfg_parser[section]['from'])
activation = cfg_parser[section]['activation']
assert activation == 'linear', 'Only linear activation supported.'
all_layers.append(Add()([all_layers[index], prev_layer]))
prev_layer = all_layers[-1]
elif section.startswith('upsample'):
stride = int(cfg_parser[section]['stride'])
assert stride == 2, 'Only stride=2 supported.'
all_layers.append(UpSampling2D(stride)(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('yolo'):
out_index.append(len(all_layers) - 1)
all_layers.append(None)
prev_layer = all_layers[-1]
elif section.startswith('net'):
pass
else:
raise ValueError(
'Unsupported section header type: {}'.format(section))
# Create and save model.
if len(out_index) == 0: out_index.append(len(all_layers) - 1)
model = Model(inputs=input_layer,
outputs=[all_layers[i] for i in out_index])
print(model.summary())
if args.weights_only:
model.save_weights('{}'.format(output_path))
print('Saved Keras weights to {}'.format(output_path))
else:
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 args.plot_model:
plot(model, to_file='{}.png'.format(output_root), show_shapes=True)
print('Saved model plot to {}.png'.format(output_root))
def _main(args):
config_path = os.path.expanduser(args.config_path)
weights_path = os.path.expanduser(args.weights_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)
output_path = os.path.expanduser(args.output_path)
assert output_path.endswith(
'.h5'), 'output path {} is not a .h5 file'.format(output_path)
output_root = os.path.splitext(output_path)[0]
# Load weights and config.
print('Loading weights.')
weights_file = open(weights_path, 'rb')
major, minor, revision = np.ndarray(
shape=(3, ), dtype='int32', buffer=weights_file.read(12))
if (major*10+minor)>=2 and major<1000 and minor<1000:
seen = np.ndarray(shape=(1,), dtype='int64', buffer=weights_file.read(8))
else:
seen = np.ndarray(shape=(1,), dtype='int32', buffer=weights_file.read(4))
print('Weights Header: ', major, minor, revision, seen)
print('Parsing Darknet config.')
unique_config_file = unique_config_sections(config_path)
cfg_parser = configparser.ConfigParser()
cfg_parser.read_file(unique_config_file)
print('Creating Keras model.')
input_layer = Input(shape=(None, None, 3))
prev_layer = input_layer
all_layers = []
weight_decay = float(cfg_parser['net_0']['decay']
) if 'net_0' in cfg_parser.sections() else 5e-4
count = 0
out_index = []
for section in cfg_parser.sections():
print('Parsing section {}'.format(section))
if section.startswith('convolutional'):
filters = int(cfg_parser[section]['filters'])
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
pad = int(cfg_parser[section]['pad'])
activation = cfg_parser[section]['activation']
batch_normalize = 'batch_normalize' in cfg_parser[section]
padding = 'same' if pad == 1 and stride == 1 else 'valid'
# Setting weights.
# Darknet serializes convolutional weights as:
# [bias/beta, [gamma, mean, variance], conv_weights]
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)
print('conv2d', 'bn'
if batch_normalize else ' ', activation, weights_shape)
conv_bias = np.ndarray(
shape=(filters, ),
dtype='float32',
buffer=weights_file.read(filters * 4))
count += filters
if batch_normalize:
bn_weights = np.ndarray(
shape=(3, filters),
dtype='float32',
buffer=weights_file.read(filters * 12))
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))
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 == 'leaky':
pass # Add advanced activation later.
elif activation != 'linear':
raise ValueError(
'Unknown activation function `{}` in section {}'.format(
activation, section))
# Create Conv2D layer
if stride>1:
# Darknet uses left and top padding instead of 'same' mode
prev_layer = ZeroPadding2D(((1,0),(1,0)))(prev_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)
if batch_normalize:
conv_layer = (BatchNormalization(
weights=bn_weight_list))(conv_layer)
prev_layer = conv_layer
if activation == 'linear':
all_layers.append(prev_layer)
elif activation == 'leaky':
act_layer = LeakyReLU(alpha=0.1)(prev_layer)
prev_layer = act_layer
all_layers.append(act_layer)
elif section.startswith('route'):
ids = [int(i) for i in cfg_parser[section]['layers'].split(',')]
layers = [all_layers[i] for i in ids]
if len(layers) > 1:
print('Concatenating route layers:', layers)
concatenate_layer = Concatenate()(layers)
all_layers.append(concatenate_layer)
prev_layer = concatenate_layer
else:
skip_layer = layers[0] # only one layer to route
all_layers.append(skip_layer)
prev_layer = skip_layer
elif section.startswith('maxpool'):
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
all_layers.append(
MaxPooling2D(
pool_size=(size, size),
strides=(stride, stride),
padding='same')(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('shortcut'):
index = int(cfg_parser[section]['from'])
activation = cfg_parser[section]['activation']
assert activation == 'linear', 'Only linear activation supported.'
all_layers.append(Add()([all_layers[index], prev_layer]))
prev_layer = all_layers[-1]
elif section.startswith('upsample'):
stride = int(cfg_parser[section]['stride'])
assert stride == 2, 'Only stride=2 supported.'
all_layers.append(UpSampling2D(stride)(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('yolo'):
out_index.append(len(all_layers)-1)
all_layers.append(None)
prev_layer = all_layers[-1]
elif section.startswith('net'):
pass
else:
raise ValueError(
'Unsupported section header type: {}'.format(section))
# Create and save model.
if len(out_index)==0: out_index.append(len(all_layers)-1)
model = Model(inputs=input_layer, outputs=[all_layers[i] for i in out_index])
print(model.summary())
if args.weights_only:
model.save_weights('{}'.format(output_path))
print('Saved Keras weights to {}'.format(output_path))
else:
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 args.plot_model:
plot(model, to_file='{}.png'.format(output_root), show_shapes=True)
print('Saved model plot to {}.png'.format(output_root))
def _main(args):
config_path = os.path.expanduser(args.config_path)
weights_path = os.path.expanduser(args.weights_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)
output_path = os.path.expanduser(args.output_path)
assert output_path.endswith(
'.h5'), 'output path {} is not a .h5 file'.format(output_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)
# TODO: Check transpose flag when implementing fully connected layers.
# transpose = (weight_header[0] > 1000) or (weight_header[1] > 1000)
print('Parsing Darknet config.')
unique_config_file = unique_config_sections(config_path)
cfg_parser = configparser.ConfigParser()
cfg_parser.read_file(unique_config_file)
print('Creating Keras model.')
if args.fully_convolutional:
image_height, image_width = None, None
else:
image_height = int(cfg_parser['net_0']['height'])
image_width = int(cfg_parser['net_0']['width'])
prev_layer = Input(shape=(image_height, image_width, 3))
all_layers = [prev_layer]
weight_decay = float(cfg_parser['net_0']['decay']
) if 'net_0' in cfg_parser.sections() else 5e-4
count = 0
for section in cfg_parser.sections():
print('Parsing section {}'.format(section))
if section.startswith('convolutional'):
filters = int(cfg_parser[section]['filters'])
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
pad = int(cfg_parser[section]['pad'])
activation = cfg_parser[section]['activation']
batch_normalize = 'batch_normalize' in cfg_parser[section]
# border_mode='same' is equivalent to Darknet pad=1
border_mode = 'same' if pad == 1 else 'valid'
# Setting weights.
# Darknet serializes convolutional weights as:
# [bias/beta, [gamma, mean, variance], conv_weights]
prev_layer_shape = K.int_shape(prev_layer)
# TODO: This assumes channel last dim_ordering.
weights_shape = (size, size, prev_layer_shape[-1], filters)
darknet_w_shape = (filters, weights_shape[2], size, size)
weights_size = np.product(weights_shape)
print('conv2d', 'bn' if batch_normalize else ' ', activation,
weights_shape)
conv_bias = np.ndarray(shape=(filters, ),
dtype='float32',
buffer=weights_file.read(filters * 4))
count += filters
if batch_normalize:
bn_weights = np.ndarray(shape=(3, filters),
dtype='float32',
buffer=weights_file.read(filters * 12))
count += 3 * filters
# TODO: Keras BatchNormalization mistakenly refers to var
# as std.
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))
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)
# TODO: Add check for Theano dim ordering.
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 == 'leaky':
pass # Add advanced activation later.
elif activation != 'linear':
raise ValueError(
'Unknown activation function `{}` in section {}'.format(
activation, section))
# Create Conv2D layer
conv_layer = (Convolution2D(
filters,
size,
size,
border_mode=border_mode,
subsample=(stride, stride),
bias=not batch_normalize,
weights=conv_weights,
activation=act_fn,
W_regularizer=l2(weight_decay)))(prev_layer)
if batch_normalize:
conv_layer = (BatchNormalization(
weights=bn_weight_list))(conv_layer)
prev_layer = conv_layer
if activation == 'linear':
all_layers.append(prev_layer)
elif activation == 'leaky':
act_layer = LeakyReLU(alpha=0.1)(prev_layer)
prev_layer = act_layer
all_layers.append(act_layer)
elif section.startswith('maxpool'):
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
all_layers.append(
MaxPooling2D(pool_size=(size, size),
strides=(stride, stride),
border_mode='same')(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('avgpool'):
if cfg_parser.items(section) != []:
raise ValueError('{} with params unsupported.'.format(section))
all_layers.append(GlobalAveragePooling2D()(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('route'):
ids = [int(i) for i in cfg_parser[section]['layers'].split(',')]
layers = [all_layers[i] for i in ids]
if len(layers) > 1:
print('Merging layers:', layers)
merge_layer = merge(layers, mode='concat')
all_layers.append(merge_layer)
prev_layer = merge_layer
else:
skip_layer = layers[0] # only one layer to route
all_layers.append(skip_layer)
prev_layer = skip_layer
elif section.startswith('reorg'):
block_size = int(cfg_parser[section]['stride'])
assert block_size == 2, 'Only reorg with stride 2 supported.'
all_layers.append(
Lambda(space_to_depth_x2,
output_shape=space_to_depth_x2_output_shape,
name='space_to_depth_x2')(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('region'):
with open('{}_anchors.txt'.format(output_root), 'w') as f:
print(cfg_parser[section]['anchors'], f) #FJE file=f
elif (section.startswith('net') or section.startswith('cost')
or section.startswith('softmax')):
pass # Configs not currently handled during model definition.
else:
raise ValueError(
'Unsupported section header type: {}'.format(section))
# Create and save model.
model = Model(input=all_layers[0], output=all_layers[-1])
print(model.summary())
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(len(remaining_weights)))
if args.plot_model:
plot(model, to_file='{}.png'.format(output_root), show_shapes=True)
print('Saved model plot to {}.png'.format(output_root))
#========= Save a sample of what you're feeding to the neural network ==========
N_sample = min(patches_imgs_train.shape[0],40)
visualize(group_images(patches_imgs_train[0:N_sample,:,:,:],5),'./'+name_experiment+'/'+"sample_input_imgs")#.show()
visualize(group_images(patches_masks_train[0:N_sample,:,:,:],5),'./'+name_experiment+'/'+"sample_input_masks")#.show()
#=========== Construct and save the model arcitecture =====
n_ch = patches_imgs_train.shape[1]
patch_height = patches_imgs_train.shape[2]
patch_width = patches_imgs_train.shape[3]
model = get_unet(n_ch, patch_height, patch_width) #the U-net model
print "Check: final output of the network:"
print model.output_shape
plot(model, to_file='./'+name_experiment+'/'+name_experiment + '_model.png') #check how the model looks like
json_string = model.to_json()
open('./'+name_experiment+'/'+name_experiment +'_architecture.json', 'w').write(json_string)
#============ Training ==================================
checkpointer = ModelCheckpoint(filepath='./'+name_experiment+'/'+name_experiment +'_best_weights.h5', verbose=1, monitor='val_categorical_accuracy', mode='max', save_best_only=True) #save at each epoch if the validation decreased
reduce_lr = ReduceLROnPlateau(monitor='val_categorical_accuracy', factor=0.1, patience=10, min_delta=0.0001, verbose=1, mode='auto')
# def step_decay(epoch):
# lrate = 0.01 #the initial learning rate (by default in keras)
# if epoch==100:
# return 0.005
# else:
def _main(args):
config_path = os.path.expanduser(args.config_path)
weights_path = os.path.expanduser(args.weights_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)
output_path = os.path.expanduser(args.output_path)
assert output_path.endswith(
'.h5'), 'output path {} is not a .h5 file'.format(output_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=(5, ), dtype='int32', buffer=weights_file.read(20))
print('Weights Header: ', weights_header)
# TODO: Check transpose flag when implementing fully connected layers.
# transpose = (weight_header[0] > 1000) or (weight_header[1] > 1000)
print('Parsing Darknet config.')
unique_config_file = unique_config_sections(config_path)
cfg_parser = configparser.ConfigParser()
cfg_parser.read_file(unique_config_file)
print('Creating Keras model.')
if args.fully_convolutional:
image_height, image_width = None, None
else:
image_height = int(cfg_parser['net_0']['height'])
image_width = int(cfg_parser['net_0']['width'])
prev_layer = Input(shape=(image_height, image_width, 3))
all_layers = [prev_layer]
outputs = []
weight_decay = float(cfg_parser['net_0']['decay']
) if 'net_0' in cfg_parser.sections() else 5e-4
count = 0
for section in cfg_parser.sections():
print('Parsing section {}'.format(section))
if section.startswith('convolutional'):
filters = int(cfg_parser[section]['filters'])
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
pad = int(cfg_parser[section]['pad'])
activation = cfg_parser[section]['activation']
batch_normalize = 'batch_normalize' in cfg_parser[section]
# Setting weights.
# Darknet serializes convolutional weights as:
# [bias/beta, [gamma, mean, variance], conv_weights]
prev_layer_shape = K.int_shape(prev_layer)
# TODO: This assumes channel last dim_ordering.
weights_shape = (size, size, prev_layer_shape[-1], filters)
darknet_w_shape = (filters, weights_shape[2], size, size)
weights_size = np.product(weights_shape)
print('conv2d', 'bn'
if batch_normalize else ' ', activation, weights_shape)
conv_bias = np.ndarray(
shape=(filters, ),
dtype='float32',
buffer=weights_file.read(filters * 4))
count += filters
if batch_normalize:
bn_weights = np.ndarray(
shape=(3, filters),
dtype='float32',
buffer=weights_file.read(filters * 12))
count += 3 * filters
# TODO: Keras BatchNormalization mistakenly refers to var
# as std.
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))
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)
# TODO: Add check for Theano dim ordering.
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 == 'leaky':
pass # Add advanced activation later.
elif activation != 'linear':
raise ValueError(
'Unknown activation function `{}` in section {}'.format(
activation, section))
padding = 'same' if pad == 1 and stride == 1 else 'valid'
# Adjust padding model for darknet.
if stride == 2:
prev_layer = ZeroPadding2D(((1, 0), (1, 0)))(prev_layer)
# 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)
if batch_normalize:
conv_layer = (BatchNormalization(
weights=bn_weight_list))(conv_layer)
prev_layer = conv_layer
if activation == 'linear':
all_layers.append(prev_layer)
elif activation == 'leaky':
act_layer = LeakyReLU(alpha=0.1)(prev_layer)
prev_layer = act_layer
all_layers.append(prev_layer)
elif section.startswith('maxpool'):
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
all_layers.append(
MaxPooling2D(
padding='same',
pool_size=(size, size),
strides=(stride, stride))(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('avgpool'):
if cfg_parser.items(section) != []:
raise ValueError('{} with params unsupported.'.format(section))
all_layers.append(GlobalAveragePooling2D()(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('route'):
ids = [int(i) for i in cfg_parser[section]['layers'].split(',')]
if len(ids) == 2:
for i, item in enumerate(ids):
if item != -1:
ids[i] = item + 1
layers = [all_layers[i] for i in ids]
if len(layers) > 1:
print('Concatenating route layers:', layers)
concatenate_layer = concatenate(layers)
all_layers.append(concatenate_layer)
prev_layer = concatenate_layer
else:
skip_layer = layers[0] # only one layer to route
all_layers.append(skip_layer)
prev_layer = skip_layer
elif section.startswith('shortcut'):
ids = [int(i) for i in cfg_parser[section]['from'].split(',')][0]
activation = cfg_parser[section]['activation']
shortcut = add([all_layers[ids], prev_layer])
if activation == 'linear':
shortcut = Activation('linear')(shortcut)
all_layers.append(shortcut)
prev_layer = all_layers[-1]
elif section.startswith('upsample'):
stride = int(cfg_parser[section]['stride'])
all_layers.append(
UpSampling2D(
size=(stride, stride))(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('yolo'):
classes = int(cfg_parser[section]['classes'])
# num = int(cfg_parser[section]['num'])
# mask = int(cfg_parser[section]['mask'])
n1, n2 = int(prev_layer.shape[1]), int(prev_layer.shape[2])
n3 = 3
n4 = (4 + 1 + classes)
yolo = Reshape((n1, n2, n3, n4))(prev_layer)
all_layers.append(yolo)
prev_layer = all_layers[-1]
outputs.append(len(all_layers) - 1)
elif (section.startswith('net')):
pass # Configs not currently handled during model definition.
else:
raise ValueError(
'Unsupported section header type: {}'.format(section))
# Create and save model.
model = Model(inputs=all_layers[0],
outputs=[all_layers[i] for i in outputs])
print(model.summary())
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))
plot(model, to_file='{}.png'.format(output_root), show_shapes=True)
print('Saved model plot to {}.png'.format(output_root))
#print(len(mobilenet_model.layers))
#print(len(mobilenet_model.output))
#mobilenet_model.save_weights('empty_mobilenet.h5')
#print("reduce model")
#model_reduced = reduce_keras_model(mobilenet_model)
#model_reduced.summary()
#print(len(model_reduced.layers))
#darknet = Model( image_input , darknet_body(image_input) )
#darknet = yolo_body(image_input, num_anchors//3, num_classes)
#plot(darknet , to_file='{}.png'.format("darknet53_yolo"), show_shapes=True)
#darknet.summary()
#darknet.save_weights('empty_darknet_body.h5')
#darknet = yolo_body(image_input, num_anchors//3, num_classes)
#plot(darknet , to_file='{}.png'.format("medium_tiny_yolo"), show_shapes=True)
#darknet.summary()
#print(len(darknet.layers))
#darknet.save_weights('empty_medium_tiny_yolo.h5')
#model = load_model("model_data/416bnfuse_small_mobilenets2_trained_model.h5")
#model.summary()
#print(len(model.layers))
#dense = DenseNet121(input_tensor=image_input,weights='imagenet')
#dense = DenseNet169(input_tensor=image_input,weights='imagenet')
dense = DenseNet201(input_tensor=image_input, weights='imagenet')
plot(dense, to_file='{}.png'.format("densenet201"), show_shapes=True)
dense.summary()
dense.save_weights('densenet_empty201.h5')
def build_own_det(img_shape=(3, 224, 224), n_classes=1000, num_priors=5):
K.set_image_dim_ordering('th')
img_input = Input(shape=img_shape)
# Steem
x = Convolution2D(64, 7, 7, border_mode='same', subsample=(1,1), name='block1_conv1', activation='relu')(img_input)
x = MaxPooling2D(pool_size=(3, 3), strides=(4, 4), padding='same', name='MaxPool_1')(x)
x = BatchNormalization(name='Batch_1')(x)
x = Convolution2D(64, 1, 1, border_mode='same', subsample=(1,1), name='block2_conv1', activation='relu')(x)
x = Convolution2D(256, 3, 3, border_mode='same', subsample=(1,1), name='block2_conv2', activation='relu')(x)
x = BatchNormalization(name='Batch_2')(x)
steem_output = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same', name='steem_output')(x)
# Inception 1
inc_1_1 = Convolution2D(64, 1, 1, border_mode='same', subsample=(1,1), name='inc_1_1', activation='relu')(steem_output)
inc_1_2_a = Convolution2D(96, 1, 1, border_mode='same', subsample=(1,1), name='inc_1_2_a', activation='relu')(steem_output)
inc_1_2_b = Convolution2D(128, 3, 3, border_mode='same', subsample=(1,1), name='inc_1_2_b', activation='relu')(inc_1_2_a)
inc_1_3_a = Convolution2D(16, 1, 1, border_mode='same', subsample=(1,1), name='inc_1_3_a', activation='relu')(steem_output)
inc_1_3_b = Convolution2D(32, 5, 5, border_mode='same', subsample=(1,1), name='inc_1_3_b', activation='relu')(inc_1_3_a)
inc_1_4_a = MaxPooling2D(pool_size=(3, 3), strides=(1, 1), padding='same', name='inc_1_4_a')(steem_output)
inc_1_4_b = Convolution2D(32, 1, 1, border_mode='same', subsample=(1,1), name='inc_1_4_b', activation='relu')(inc_1_4_a)
out_inc_1 = merge([inc_1_1, inc_1_2_b, inc_1_3_b, inc_1_4_b], mode='concat', concat_axis=1, name='out_inc_1')
# Residual 1
add_r1 = Add(name='add_r1')([steem_output, out_inc_1])
add_1 = Convolution2D(480, 1, 1, border_mode='same', subsample=(1,1), name='add_1', activation='relu')(add_r1)
# Inception 2
inc_2_1 = Convolution2D(128, 1, 1, border_mode='same', subsample=(1,1), name='inc_2_1', activation='relu')(add_1)
inc_2_2_a = Convolution2D(128, 1, 1, border_mode='same', subsample=(1,1), name='inc_2_2_a', activation='relu')(add_1)
inc_2_2_b = Convolution2D(192, 3, 3, border_mode='same', subsample=(1,1), name='inc_2_2_b', activation='relu')(inc_2_2_a)
inc_2_3_a = Convolution2D(32, 1, 1, border_mode='same', subsample=(1,1), name='inc_2_3_a', activation='relu')(add_1)
inc_2_3_b = Convolution2D(96, 5, 5, border_mode='same', subsample=(1,1), name='inc_2_3_b', activation='relu')(inc_2_3_a)
inc_2_4_a = MaxPooling2D(pool_size=(3, 3), strides=(1, 1), border_mode='same', name='inc_2_4_a')(add_1)
inc_2_4_b = Convolution2D(64, 1, 1, border_mode='same', subsample=(1,1), name='inc_2_4_b', activation='relu')(inc_2_4_a)
out_inc_2 = merge([inc_2_1, inc_2_2_b, inc_2_3_b, inc_2_4_b], mode='concat', concat_axis=1, name='out_inc_2')
# Residual 2
add_2 = Add(name='add_2')([add_1, out_inc_2])
# Parallel block
out_inc = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same', name='out_inc')(add_2)
conv_3 = YOLOConvolution2D(512, 3, 3, border_mode='same', subsample=(1,1), name='block3_conv1')(out_inc)
conv_3 = LeakyReLU(alpha=0.1)(conv_3)
conv_4 = YOLOConvolution2D(1024, 3, 3, border_mode='same', subsample=(1,1), name='block4_conv1')(conv_3)
conv_4 = LeakyReLU(alpha=0.1)(conv_4)
conv_5 = YOLOConvolution2D(512, 3, 3, border_mode='same', subsample=(1,1), name='block5_conv1')(conv_4)
conv_5 = LeakyReLU(alpha=0.1)(conv_5)
conv_6 = YOLOConvolution2D(1024, 3, 3, border_mode='same', subsample=(1,1), name='block6_conv1')(conv_5)
conv_6 = LeakyReLU(alpha=0.1)(conv_6)
conv_7 = YOLOConvolution2D(512, 3, 3, border_mode='same', subsample=(1,1), name='block7_conv1')(conv_6)
conv_7 = LeakyReLU(alpha=0.1)(conv_7)
conv_8 = YOLOConvolution2D(1024, 3, 3, border_mode='same', subsample=(1,1), name='block8_conv1')(conv_7)
conv_8 = LeakyReLU(alpha=0.1)(conv_8)
reorg = (Reorg())(add_2)
concat = Concatenate(axis=1)([conv_8, reorg])
conv_9 = Convolution2D(1024, 3, 3, border_mode='same', subsample=(1,1), name='block9_conv1', activation='relu')(concat)
last_conv = Convolution2D(num_priors * (4 + n_classes + 1), (1, 1), padding='same', strides=(1, 1), activation='relu')(conv_9)
model = Model(input=img_input, output=last_conv)
plot(model, to_file='team7_model_det.png', show_shapes=True, show_layer_names=True)
return model
print('validation set size %d' % steerings_test.shape[0])
# check training data ok
paths_train = np.array(paths_train)
steerings_train = np.array(steerings_train)
assert paths_train.shape[0] == steerings_train.shape[0]
print('training set size %d' % paths_train.shape[0])
print('Creating model...')
model = model_builder()
plot(model, to_file='model.png', show_shapes=True, show_layer_names=False)
# Train model
print('Validating traing / testing data size ...')
assert paths_train.shape[0] == steerings_train.shape[0]
assert paths_test.shape[0] == steerings_test.shape[0]
print('Data looks good!')
train_size = paths_train.shape[0]
test_size = paths_test.shape[0]
batch_size = 128
nb_epochs = 10
print('Start training... batch size %d' % batch_size)
train_generator = batches(
paths_train, steerings_train, batch_size=batch_size, training=True)
def _main(args):
config_path = os.path.expanduser(args.config_path)
weights_path = os.path.expanduser(args.weights_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)
output_path = os.path.expanduser(args.output_path)
assert output_path.endswith(
'.h5'), 'output path {} is not a .h5 file'.format(output_path)
output_root = os.path.splitext(output_path)[0]
# Load weights and config.
print('Loading weights.')
weights_file = open(weights_path, 'rb')
major, minor, revision = np.ndarray(
shape=(3, ), dtype='int32', buffer=weights_file.read(12))
if (major*10+minor)>=2 and major<1000 and minor<1000:
seen = np.ndarray(shape=(1,), dtype='int64', buffer=weights_file.read(8))
else:
seen = np.ndarray(shape=(1,), dtype='int32', buffer=weights_file.read(4))
print('Weights Header: ', major, minor, revision, seen)
print('Parsing Darknet config.')
# 读取yolov3最开始的配置文件，将其中重名的section添加序号修改为不同名称
# 该函数返回一个可迭代的object
unique_config_file = unique_config_sections(config_path)
cfg_parser = configparser.ConfigParser()
# 通过read_file（）读取file-like形式的object配置文件，就是上述修改section名称之后的文件
cfg_parser.read_file(unique_config_file)
print('Creating Keras model.')
input_layer = Input(shape=(None, None, 3))
prev_layer = input_layer
all_layers = []
weight_decay = float(cfg_parser['net_0']['decay']
) if 'net_0' in cfg_parser.sections() else 5e-4
count = 0
out_index = []
for section in cfg_parser.sections():
print('Parsing section {}'.format(section))
if section.startswith('convolutional'):
filters = int(cfg_parser[section]['filters'])
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
pad = int(cfg_parser[section]['pad'])
activation = cfg_parser[section]['activation']
batch_normalize = 'batch_normalize' in cfg_parser[section]
padding = 'same' if pad == 1 and stride == 1 else 'valid'
# Setting weights.
# Darknet serializes convolutional weights as:
# [bias/beta, [gamma, mean, variance], conv_weights]
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)
print('conv2d', 'bn'
if batch_normalize else ' ', activation, weights_shape)
conv_bias = np.ndarray(
shape=(filters, ),
dtype='float32',
buffer=weights_file.read(filters * 4))
count += filters
if batch_normalize:
bn_weights = np.ndarray(
shape=(3, filters),
dtype='float32',
buffer=weights_file.read(filters * 12))
count += 3 * filters
bn_weight_list = [
bn_weights[0], # scale gamma
conv_bias, # shift beta，这个不是第一个卷积层的bias吗？为什么当做BN层的beta
bn_weights[1], # running mean，训练的时候BN层均值和方差是随样本变化的，但是在测试的时候，
# 单样本输入，均值和方差用的是整个训练集在每个BN层的均值和方差
bn_weights[2] # running var
]
conv_weights = np.ndarray(
shape=darknet_w_shape,
dtype='float32',
buffer=weights_file.read(weights_size * 4))
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
] #也就是说，有BN层的时候卷积核是没有bias的，conv_bias存储的是BN层的beta值
# Handle activation.
act_fn = None
if activation == 'leaky':
pass # Add advanced activation later.
elif activation != 'linear':
raise ValueError(
'Unknown activation function `{}` in section {}'.format(
activation, section))
# Create Conv2D layer
if stride>1: # 该网络中除了stride=1的步长，剩下的就是stride=2的步长，也就是下采样2倍
# Darknet uses left and top padding instead of 'same' mode
prev_layer = ZeroPadding2D(((1,0),(1,0)))(prev_layer)
conv_layer = (Conv2D(
filters, (size, size),
strides=(stride, stride),
kernel_regularizer=l2(weight_decay),
use_bias=not batch_normalize,
weights=conv_weights, # 在层层调用的init函数的**kwargs中找到weights关键字，将其初始化为卷积核权重
activation=act_fn,
padding=padding))(prev_layer)
if batch_normalize:
conv_layer = (BatchNormalization(
weights=bn_weight_list))(conv_layer)
prev_layer = conv_layer
if activation == 'linear':
all_layers.append(prev_layer)
elif activation == 'leaky':
act_layer = LeakyReLU(alpha=0.1)(prev_layer) #leakyReLu函数，alpha表示小于0部分的斜率
prev_layer = act_layer
all_layers.append(act_layer)
elif section.startswith('route'): # 表示concatenate结构，两层并联在一起输出的层
# 在配置文件中route有两种，一种是只有单层layer的，一个是有两层layer的，
# 单层layer的表示该层已经到达输出层，需要向上回溯若干层，再做上采样，用于与其它层concatenate然后在另一个尺寸输出
# 两侧layer的表示该层属于concatenate层，这两层需要并联在一起输出
ids = [int(i) for i in cfg_parser[section]['layers'].split(',')]
layers = [all_layers[i] for i in ids]
if len(layers) > 1:
print('Concatenating route layers:', layers)
concatenate_layer = Concatenate()(layers)
all_layers.append(concatenate_layer)
prev_layer = concatenate_layer
else:
skip_layer = layers[0] # only one layer to route
all_layers.append(skip_layer)
prev_layer = skip_layer
elif section.startswith('maxpool'):
size = int(cfg_parser[section]['size'])
stride = int(cfg_parser[section]['stride'])
all_layers.append(
MaxPooling2D(
pool_size=(size, size),
strides=(stride, stride),
padding='same')(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('shortcut'): # shortcut表示残差结构中的add层，将当前层与与2层前相加
index = int(cfg_parser[section]['from'])
activation = cfg_parser[section]['activation']
assert activation == 'linear', 'Only linear activation supported.'
all_layers.append(Add()([all_layers[index], prev_layer]))
prev_layer = all_layers[-1]
elif section.startswith('upsample'):
stride = int(cfg_parser[section]['stride'])
assert stride == 2, 'Only stride=2 supported.'
all_layers.append(UpSampling2D(stride)(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith('yolo'):
out_index.append(len(all_layers)-1)
all_layers.append(None)
prev_layer = all_layers[-1]
elif section.startswith('net'):
pass
else:
raise ValueError(
'Unsupported section header type: {}'.format(section))
# Create and save model.
if len(out_index)==0: out_index.append(len(all_layers)-1)
model = Model(inputs=input_layer, outputs=[all_layers[i] for i in out_index]) # out_index是三个输出层序号
print(model.summary())
if args.weights_only:
model.save_weights('{}'.format(output_path))
print('Saved Keras weights to {}'.format(output_path))
else:
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 args.plot_model:
plot(model, to_file='{}.png'.format(output_root), show_shapes=True)
print('Saved model plot to {}.png'.format(output_root))
def _main(config_path, weights_path, output_path, weights_only, plot_model):
assert os.path.isfile(config_path), 'missing "%s"' % config_path
assert os.path.isfile(weights_path), 'missing "%s"' % weights_path
assert config_path.endswith('.cfg'), \
'"%s" is not a .cfg file' % os.path.basename(config_path)
assert weights_path.endswith('.weights'), \
'"%s" is not a .weights file' % os.path.basename(config_path)
output_dir = update_path(os.path.dirname(output_path))
assert os.path.isdir(output_dir), 'missing "%s"' % output_dir
output_path = os.path.join(output_dir, os.path.basename(output_path))
assert output_path.endswith('.h5'), \
'output path "%s" is not a .h5 file' % os.path.basename(output_path)
# Load weights and config.
logging.info('Loading weights: %s', weights_path)
weights_file = open(weights_path, 'rb')
major, minor, revision = np.ndarray(shape=(3, ),
dtype='int32',
buffer=weights_file.read(12))
if (major * 10 + minor) >= 2 and major < 1000 and minor < 1000:
seen = np.ndarray(shape=(1, ),
dtype='int64',
buffer=weights_file.read(8))
else:
seen = np.ndarray(shape=(1, ),
dtype='int32',
buffer=weights_file.read(4))
logging.info('Weights Header: %i.%i.%i %s', major, minor, revision,
repr(seen.tolist()))
logging.info('Parsing Darknet config: %s', config_path)
unique_config_file = unique_config_sections(config_path)
cfg_parser = configparser.ConfigParser()
cfg_parser.read_file(unique_config_file)
logging.info('Creating Keras model.')
input_layer = Input(shape=(None, None, 3))
prev_layer = input_layer
all_layers = []
weight_decay = float(cfg_parser['net_0']['decay']
) if 'net_0' in cfg_parser.sections() else 5e-4
count = 0
out_index = []
for section in tqdm.tqdm(cfg_parser.sections()):
logging.info('Parsing section "%s"', section)
(all_layers, cfg_parser, section, prev_layer, weights_file, count,
weight_decay, out_index) = parse_section(all_layers, cfg_parser,
section, prev_layer,
weights_file, count,
weight_decay, out_index)
# Create and save model.
if len(out_index) == 0:
out_index.append(len(all_layers) - 1)
model = Model(inputs=input_layer,
outputs=[all_layers[i] for i in out_index])
logging.info(model.summary(line_length=120))
if weights_only:
model.save_weights(output_path)
logging.info('Saved Keras weights to "%s"', output_path)
else:
model.save(output_path)
logging.info('Saved Keras model to "%s"', output_path)
# Check to see if all weights have been read.
remaining_weights = len(weights_file.read()) / 4
weights_file.close()
logging.info('Read %i of %i from Darknet weight.', count,
count + remaining_weights)
if remaining_weights > 0:
logging.warning('there are %i unused weights', remaining_weights)
if plot_model:
path_img = '%s.png' % os.path.splitext(output_path)[0]
plot(model, to_file=path_img, show_shapes=True)
logging.info('Saved model plot to %s', path_img)
def _main(args):
config_path = os.path.expanduser(args.config_path)
weights_path = os.path.expanduser(args.weights_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)
output_path = os.path.expanduser(args.output_path)
assert output_path.endswith(
".h5"), "output path {} is not a .h5 file".format(output_path)
output_root = os.path.splitext(output_path)[0]
# Load weights and config.
print("Loading weights.")
weights_file = open(weights_path, "rb")
major, minor, revision = np.ndarray(shape=(3, ),
dtype="int32",
buffer=weights_file.read(12))
if (major * 10 + minor) >= 2 and major < 1000 and minor < 1000:
seen = np.ndarray(shape=(1, ),
dtype="int64",
buffer=weights_file.read(8))
else:
seen = np.ndarray(shape=(1, ),
dtype="int32",
buffer=weights_file.read(4))
print("Weights Header: ", major, minor, revision, seen)
print("Parsing Darknet config.")
unique_config_file = unique_config_sections(config_path)
cfg_parser = configparser.ConfigParser()
cfg_parser.read_file(unique_config_file)
print("Creating Keras model.")
input_layer = Input(shape=(None, None, 3))
prev_layer = input_layer
all_layers = []
weight_decay = (float(cfg_parser["net_0"]["decay"])
if "net_0" in cfg_parser.sections() else 5e-4)
count = 0
out_index = []
for section in cfg_parser.sections():
print("Parsing section {}".format(section))
if section.startswith("convolutional"):
filters = int(cfg_parser[section]["filters"])
size = int(cfg_parser[section]["size"])
stride = int(cfg_parser[section]["stride"])
pad = int(cfg_parser[section]["pad"])
activation = cfg_parser[section]["activation"]
batch_normalize = "batch_normalize" in cfg_parser[section]
padding = "same" if pad == 1 and stride == 1 else "valid"
# Setting weights.
# Darknet serializes convolutional weights as:
# [bias/beta, [gamma, mean, variance], conv_weights]
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)
print("conv2d", "bn" if batch_normalize else " ", activation,
weights_shape)
conv_bias = np.ndarray(shape=(filters, ),
dtype="float32",
buffer=weights_file.read(filters * 4))
count += filters
if batch_normalize:
bn_weights = np.ndarray(
shape=(3, filters),
dtype="float32",
buffer=weights_file.read(filters * 12),
)
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),
)
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 == "leaky":
pass # Add advanced activation later.
elif activation != "linear":
raise ValueError(
"Unknown activation function `{}` in section {}".format(
activation, section))
# Create Conv2D layer
if stride > 1:
# Darknet uses left and top padding instead of 'same' mode
prev_layer = ZeroPadding2D(((1, 0), (1, 0)))(prev_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)
if batch_normalize:
conv_layer = (BatchNormalization(
weights=bn_weight_list))(conv_layer)
prev_layer = conv_layer
if activation == "linear":
all_layers.append(prev_layer)
elif activation == "leaky":
act_layer = LeakyReLU(alpha=0.1)(prev_layer)
prev_layer = act_layer
all_layers.append(act_layer)
elif section.startswith("route"):
ids = [int(i) for i in cfg_parser[section]["layers"].split(",")]
layers = [all_layers[i] for i in ids]
if len(layers) > 1:
print("Concatenating route layers:", layers)
concatenate_layer = Concatenate()(layers)
all_layers.append(concatenate_layer)
prev_layer = concatenate_layer
else:
skip_layer = layers[0] # only one layer to route
all_layers.append(skip_layer)
prev_layer = skip_layer
elif section.startswith("maxpool"):
size = int(cfg_parser[section]["size"])
stride = int(cfg_parser[section]["stride"])
all_layers.append(
MaxPooling2D(pool_size=(size, size),
strides=(stride, stride),
padding="same")(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith("shortcut"):
index = int(cfg_parser[section]["from"])
activation = cfg_parser[section]["activation"]
assert activation == "linear", "Only linear activation supported."
all_layers.append(Add()([all_layers[index], prev_layer]))
prev_layer = all_layers[-1]
elif section.startswith("upsample"):
stride = int(cfg_parser[section]["stride"])
assert stride == 2, "Only stride=2 supported."
all_layers.append(UpSampling2D(stride)(prev_layer))
prev_layer = all_layers[-1]
elif section.startswith("yolo"):
out_index.append(len(all_layers) - 1)
all_layers.append(None)
prev_layer = all_layers[-1]
elif section.startswith("net"):
pass
else:
raise ValueError(
"Unsupported section header type: {}".format(section))
# Create and save model.
if len(out_index) == 0:
out_index.append(len(all_layers) - 1)
model = Model(inputs=input_layer,
outputs=[all_layers[i] for i in out_index])
print(model.summary())
if args.weights_only:
model.save_weights("{}".format(output_path))
print("Saved Keras weights to {}".format(output_path))
else:
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(
f"Read {count:0.0f} of {count + remaining_weights:0.0f} from Darknet weights."
)
if remaining_weights > 0:
print("Warning: {} unused weights".format(remaining_weights))
if args.plot_model:
plot(model, to_file="{}.png".format(output_root), show_shapes=True)
print("Saved model plot to {}.png".format(output_root))
img_size= (32,32)
img_out_dir="segmentation_results".format(dataset)
model_out_dir="{}/model_output_result".format(dataset)
auc_out_dir="auc".format(dataset)
train_dir="F:/my model/data/DRIVE/training/".format(dataset)
test_dir="F:/my model/data/DRIVE/test/".format(dataset)
if not os.path.isdir(img_out_dir):
os.makedirs(img_out_dir)
if not os.path.isdir(model_out_dir):
os.makedirs(model_out_dir)
if not os.path.isdir(auc_out_dir):
os.makedirs(auc_out_dir)
model = Unet_model(img_size, filter_size,init_lr)
model.summary()
plot(model,model_out_dir+'model.png')
# set training and validation dataset
train_imgs, train_vessels =getdata.ready_image(train_dir, img_size=img_size, dataset=dataset)
train_vessels=np.expand_dims(train_vessels, axis=3)
# set test dataset
test_imgs, test_vessels, test_masks=getdata.ready_image(test_dir, img_size=img_size, dataset=dataset, mask=True)
checkpointer = ModelCheckpoint(filepath= model_out_dir+'_best_weights.h5', verbose=1, monitor='val_loss',
mode='auto', save_best_only=True)
model.fit(train_imgs, train_vessels, epochs=N_epochs, batch_size=batch_size, validation_split=0.1, callbacks=[checkpointer])
##for n_round in range(n_rounds):
def plot_neural_net(neuralnet, filename='net-visualzation.png'):
plot(neuralnet, to_file=filename, show_shapes=True, show_layer_names=True)