def _tiny_res_block(inputs, in_channels, data_format):
net = _conv2d_fixed_padding(inputs, in_channels, kernel_size=3)
route = net
#_,split=tf.split(net,num_or_size_splits=2,axis=1 if data_format =="NCHW" else 3)
split = net[:, int(
in_channels /
2):, :, :] if data_format == "NCHW" else net[:, :, :,
int(in_channels / 2):]
net = _conv2d_fixed_padding(split, in_channels / 2, kernel_size=3)
route1 = net
net = _conv2d_fixed_padding(net, in_channels / 2, kernel_size=3)
net = tf.concat([net, route1], axis=1 if data_format == 'NCHW' else 3)
net = _conv2d_fixed_padding(net, in_channels, kernel_size=1)
feat = net
net = tf.concat([net, route], axis=1 if data_format == 'NCHW' else 3)
net = slim.max_pool2d(net, [2, 2], scope='pool2')
return net, feat
def yolo_v3_tiny(inputs, num_classes, is_training=False, data_format='NCHW', reuse=False):
"""
Creates YOLO v3 tiny model.
:param inputs: a 4-D tensor of size [batch_size, height, width, channels].
Dimension batch_size may be undefined. The channel order is RGB.
:param num_classes: number of predicted classes.
:param is_training: whether is training or not.
:param data_format: data format NCHW or NHWC.
:param reuse: whether or not the network and its variables should be reused.
:return:
"""
# it will be needed later on
img_size = inputs.get_shape().as_list()[1:3]
# transpose the inputs to NCHW
if data_format == 'NCHW':
inputs = tf.transpose(inputs, [0, 3, 1, 2])
# normalize values to range [0..1]
inputs = inputs / 255
# set batch norm params
batch_norm_params = {
'decay': _BATCH_NORM_DECAY,
'epsilon': _BATCH_NORM_EPSILON,
'scale': True,
'is_training': is_training,
'fused': None, # Use fused batch norm if possible.
}
# Set activation_fn and parameters for conv2d, batch_norm.
with slim.arg_scope([slim.conv2d, slim.batch_norm, _fixed_padding, slim.max_pool2d], data_format=data_format):
with slim.arg_scope([slim.conv2d, slim.batch_norm, _fixed_padding], reuse=reuse):
with slim.arg_scope([slim.conv2d],
normalizer_fn=slim.batch_norm,
normalizer_params=batch_norm_params,
biases_initializer=None,
activation_fn=lambda x: tf.nn.relu(x)):
with tf.variable_scope('yolo-v3-tiny'):
for i in range(6):
inputs = _conv2d_fixed_padding(
inputs, 16 * pow(2, i), 3)
if i == 4:
route_1 = inputs
if i == 5:
inputs = slim.max_pool2d(
inputs, [2, 2], stride=1, padding="SAME", scope='pool2')
else:
inputs = slim.max_pool2d(
inputs, [2, 2], scope='pool2')
inputs = _conv2d_fixed_padding(inputs, 1024, 3)
inputs = _conv2d_fixed_padding(inputs, 256, 1)
route_2 = inputs
inputs = _conv2d_fixed_padding(inputs, 512, 3)
# inputs = _conv2d_fixed_padding(inputs, 255, 1)
detect_1 = _detection_layer(
inputs, num_classes, _ANCHORS[3:6], img_size, data_format)
detect_1 = tf.identity(detect_1, name='detect_1')
inputs = _conv2d_fixed_padding(route_2, 128, 1)
upsample_size = route_1.get_shape().as_list()
inputs = _upsample(inputs, upsample_size, data_format)
inputs = tf.concat([inputs, route_1],
axis=1 if data_format == 'NCHW' else 3)
inputs = _conv2d_fixed_padding(inputs, 256, 3)
# inputs = _conv2d_fixed_padding(inputs, 255, 1)
detect_2 = _detection_layer(
inputs, num_classes, _ANCHORS[0:3], img_size, data_format)
detect_2 = tf.identity(detect_2, name='detect_2')
detections = tf.concat([detect_1, detect_2], axis=1)
detections = tf.identity(detections, name='detections')
return detections
def yolo_v3_tiny(inputs, num_classes, is_training=False, data_format='NCHW', reuse=False):
# it will be needed later on
img_size = inputs.get_shape().as_list()[1:3]
# transpose the inputs to NCHW
if data_format == 'NCHW':
inputs = tf.transpose(inputs, [0, 3, 1, 2])
# normalize values to range [0..1]
inputs = inputs / 255
# set batch norm params
batch_norm_params = {
'decay': _BATCH_NORM_DECAY,
'epsilon': _BATCH_NORM_EPSILON,
'scale': True,
'is_training': is_training,
'fused': None, # Use fused batch norm if possible.
}
# Set activation_fn and parameters for conv2d, batch_norm.
with slim.arg_scope([slim.conv2d, slim.batch_norm, _fixed_padding, slim.max_pool2d], data_format=data_format):
with slim.arg_scope([slim.conv2d, slim.batch_norm, _fixed_padding], reuse=reuse):
with slim.arg_scope([slim.conv2d],
normalizer_fn=slim.batch_norm,
normalizer_params=batch_norm_params,
biases_initializer=None,
activation_fn=lambda x: tf.nn.leaky_relu(x, alpha=_LEAKY_RELU)):
#with tf.variable_scope('yolo-v3-tiny'):
for i in range(6):
inputs = _conv2d_fixed_padding(
inputs, 16 * pow(2, i), 3)
if i == 4:
route_1 = inputs
if i == 5:
inputs = slim.max_pool2d(
inputs, [2, 2], stride=1, padding="SAME", scope='pool2')
else:
inputs = slim.max_pool2d(
inputs, [2, 2], scope='pool2')
inputs = _conv2d_fixed_padding(inputs, 1024, 3)
inputs = _conv2d_fixed_padding(inputs, 256, 1)
route_2 = inputs
inputs = _conv2d_fixed_padding(inputs, 512, 3)
# inputs = _conv2d_fixed_padding(inputs, 255, 1)
detect_1 = _detection_layer(
inputs, num_classes, _ANCHORS[3:6], img_size, data_format)
detect_1 = tf.identity(detect_1, name='detect_1')
inputs = _conv2d_fixed_padding(route_2, 128, 1)
upsample_size = route_1.get_shape().as_list()
inputs = _upsample(inputs, upsample_size, data_format)
inputs = tf.concat([inputs, route_1],
axis=1 if data_format == 'NCHW' else 3)
inputs = _conv2d_fixed_padding(inputs, 256, 3)
# inputs = _conv2d_fixed_padding(inputs, 255, 1)
detect_2 = _detection_layer(
inputs, num_classes, _ANCHORS[0:3], img_size, data_format)
detect_2 = tf.identity(detect_2, name='detect_2')
detections = tf.concat([detect_1, detect_2], axis=1)
detections = tf.identity(detections, name='detections')
return detections
def yolo_v3_tiny_pan(inputs, num_classes, is_training=False, data_format='NCHW', reuse=False):
"""
Creates YOLO v3 tiny pan model.
:param inputs: a 4-D tensor of size [batch_size, height, width, channels].
Dimension batch_size may be undefined. The channel order is RGB.
:param num_classes: number of predicted classes.
:param is_training: whether is training or not.
:param data_format: data format NCHW or NHWC.
:param reuse: whether or not the network and its variables should be reused.
:return:
"""
_ANCHORS = [(10,13), (16,30), (33,23), (30,61), (62,45), (59,119), (116,90), (156,198), (373,326)]
# it will be needed later on
img_size = inputs.get_shape().as_list()[1:3]
# transpose the inputs to NCHW
if data_format == 'NCHW':
inputs = tf.transpose(inputs, [0, 3, 1, 2])
# normalize values to range [0..1]
inputs = inputs / 255
# set batch norm params
batch_norm_params = {
'decay': _BATCH_NORM_DECAY,
'epsilon': _BATCH_NORM_EPSILON,
'scale': True,
'is_training': is_training,
'fused': None, # Use fused batch norm if possible.
}
# Set activation_fn and parameters for conv2d, batch_norm.
with slim.arg_scope([slim.conv2d, slim.batch_norm, _fixed_padding, slim.max_pool2d], data_format=data_format):
with slim.arg_scope([slim.conv2d, slim.batch_norm, _fixed_padding], reuse=reuse):
with slim.arg_scope([slim.conv2d],
normalizer_fn=slim.batch_norm,
normalizer_params=batch_norm_params,
biases_initializer=None,
activation_fn=lambda x: tf.nn.leaky_relu(x, alpha=_LEAKY_RELU)):
with tf.variable_scope('yolo-v3-tiny-pan'):
# 0 conv 16 3 x 3/ 1 544 x 544 x 3 -> 544 x 544 x 16 0.256 BF
inputs = _conv2d_fixed_padding(inputs, 16, 3)
# 1 max 2 x 2/ 2 544 x 544 x 16 -> 272 x 272 x 16 0.005 BF
inputs = slim.max_pool2d( inputs, [2, 2], scope='pool2')
route_1 = inputs
#2 conv 32 3 x 3/ 1 272 x 272 x 16 -> 272 x 272 x 32 0.682 BF
inputs = _conv2d_fixed_padding(inputs, 16, 3)
#3 max 2 x 2/ 2 272 x 272 x 32 -> 136 x 136 x 32 0.002 BF
inputs = slim.max_pool2d( inputs, [2, 2], scope='pool2')
route_3 = inputs
#4 conv 64 3 x 3/ 1 136 x 136 x 32 -> 136 x 136 x 64 0.682 BF
inputs = _conv2d_fixed_padding(inputs, 16, 3)
#5 max 2 x 2/ 2 136 x 136 x 64 -> 68 x 68 x 64 0.001 BF
inputs = slim.max_pool2d( inputs, [2, 2], scope='pool2')
route_5 = inputs
#6 conv 128 3 x 3/ 1 68 x 68 x 64 -> 68 x 68 x 128 0.682 BF
inputs = _conv2d_fixed_padding(inputs, 16, 3)
route_6 = inputs
#7 max 2 x 2/ 2 68 x 68 x 128 -> 34 x 34 x 128 0.001 BF
inputs = slim.max_pool2d( inputs, [2, 2], scope='pool2')
route_7 = inputs
#8 conv 256 3 x 3/ 1 34 x 34 x 128 -> 34 x 34 x 256 0.682 BF
inputs = _conv2d_fixed_padding(inputs, 16, 3)
route_8 = inputs
#9 max 2 x 2/ 2 34 x 34 x 256 -> 17 x 17 x 256 0.000 BF
inputs = slim.max_pool2d( inputs, [2, 2], scope='pool2')
route_9 = inputs
#10 conv 512 3 x 3/ 1 17 x 17 x 256 -> 17 x 17 x 512 0.682 BF
inputs = _conv2d_fixed_padding(inputs, 16, 3)
#11 max 2 x 2/ 1 17 x 17 x 512 -> 17 x 17 x 512 0.001 BF
inputs = slim.max_pool2d( inputs, [2, 2], stride=1, padding="SAME", scope='pool2')
#12 conv 1024 3 x 3/ 1 17 x 17 x 512 -> 17 x 17 x1024 2.727 BF
inputs = _conv2d_fixed_padding(inputs, 1024, 3)
#13 conv 256 1 x 1/ 1 17 x 17 x1024 -> 17 x 17 x 256 0.152 BF
inputs = _conv2d_fixed_padding(inputs, 256, 1)
#14 conv 512 3 x 3/ 1 17 x 17 x 256 -> 17 x 17 x 512 0.682 BF
inputs = _conv2d_fixed_padding(inputs, 512, 3)
route_14 = inputs
#15 conv 128 1 x 1/ 1 17 x 17 x 512 -> 17 x 17 x 128 0.038 BF
inputs = _conv2d_fixed_padding(inputs, 128, 1)
#16 upsample 2x 17 x 17 x 128 -> 34 x 34 x 128
inputs = _upsample(inputs, route_8.get_shape().as_list(), data_format)
#17 route 16 8
inputs = tf.concat([inputs, route_8], axis=1 if data_format == 'NCHW' else 3)
#18 conv 128 1 x 1/ 1 34 x 34 x 384 -> 34 x 34 x 128 0.114 BF
inputs = _conv2d_fixed_padding(inputs, 128, 1)
#19 conv 256 3 x 3/ 1 34 x 34 x 128 -> 34 x 34 x 256 0.682 BF
inputs = _conv2d_fixed_padding(inputs, 256, 3)
route_19 = inputs
#20 conv 128 1 x 1/ 1 34 x 34 x 256 -> 34 x 34 x 128 0.076 BF
inputs = _conv2d_fixed_padding(inputs, 128, 1)
#21 upsample 2x 34 x 34 x 128 -> 68 x 68 x 128
inputs = _upsample(inputs, route_6.get_shape().as_list(), data_format)
#22 route 21 6
inputs = tf.concat([inputs, route_6], axis=1 if data_format == 'NCHW' else 3)
#23 conv 64 1 x 1/ 1 68 x 68 x 256 -> 68 x 68 x 64 0.152 BF
inputs = _conv2d_fixed_padding(inputs, 64, 1)
#24 conv 128 3 x 3/ 1 68 x 68 x 64 -> 68 x 68 x 128 0.682 BF
inputs = _conv2d_fixed_padding(inputs, 128, 3)
route_24 = inputs
#25 route 1
inputs = route_1
#26 reorg / 2 272 x 272 x 16 -> 136 x 136 x 64
inputs = _reorg(inputs, 2)
#27 route 3 26
inputs = tf.concat([route_3, inputs], axis=1 if data_format == 'NCHW' else 3)
#28 reorg / 2 136 x 136 x 96 -> 68 x 68 x 384
inputs = _reorg(inputs, 2)
#29 route 5 28
inputs = tf.concat([route_5, inputs], axis=1 if data_format == 'NCHW' else 3)
#30 reorg / 2 68 x 68 x 448 -> 34 x 34 x1792
inputs = _reorg(inputs, 2)
#31 route 7 30
inputs = tf.concat([route_7, inputs], axis=1 if data_format == 'NCHW' else 3)
#32 reorg / 2 34 x 34 x1920 -> 17 x 17 x7680
inputs = _reorg(inputs, 2)
#33 route 9 32
inputs = tf.concat([route_9, inputs], axis=1 if data_format == 'NCHW' else 3)
route_33 = inputs
#34 conv 64 1 x 1/ 1 17 x 17 x7936 -> 17 x 17 x 64 0.294 BF
inputs = _conv2d_fixed_padding(inputs, 64, 1)
#35 upsample 4x 17 x 17 x 64 -> 68 x 68 x 64
inputs = _upsample(inputs, route_24.get_shape().as_list(), data_format)
#36 route 35 24
inputs = tf.concat([inputs, route_24], axis=1 if data_format == 'NCHW' else 3)
#37 conv 128 3 x 3/ 1 68 x 68 x 192 -> 68 x 68 x 128 2.046 BF
inputs = _conv2d_fixed_padding(inputs, 128, 3)
#38 conv 18 1 x 1/ 1 68 x 68 x 128 -> 68 x 68 x 18 0.021 BF
inputs = _conv2d_fixed_padding(inputs, 18, 1)
#39 yolo
detect_1 = _detection_layer(inputs, num_classes, _ANCHORS[0:3], img_size, data_format)
detect_1 = tf.identity(detect_1, name='detect_1')
#40 route 33
inputs = route_33
#41 conv 128 1 x 1/ 1 17 x 17 x7936 -> 17 x 17 x 128 0.587 BF
inputs = _conv2d_fixed_padding(inputs, 128, 1)
#42 upsample 2x 17 x 17 x 128 -> 34 x 34 x 128
inputs = _upsample(inputs, route_19.get_shape().as_list(), data_format)
#43 route 42 19
inputs = tf.concat([inputs, route_19], axis=1 if data_format == 'NCHW' else 3)
#44 conv 256 3 x 3/ 1 34 x 34 x 384 -> 34 x 34 x 256 2.046 BF
inputs = _conv2d_fixed_padding(inputs, 256, 3)
#45 conv 18 1 x 1/ 1 34 x 34 x 256 -> 34 x 34 x 18 0.011 BF
inputs = _conv2d_fixed_padding(inputs, 18, 1)
#46 yolo
detect_2 = _detection_layer(inputs, num_classes, _ANCHORS[3:6], img_size, data_format)
detect_2 = tf.identity(detect_2, name='detect_2')
#47 route 33
inputs = route_33
#48 conv 256 1 x 1/ 1 17 x 17 x7936 -> 17 x 17 x 256 1.174 BF
inputs = _conv2d_fixed_padding(inputs, 256, 1)
#49 route 48 14
inputs = tf.concat([inputs, route_14], axis=1 if data_format == 'NCHW' else 3)
#50 conv 512 3 x 3/ 1 17 x 17 x 768 -> 17 x 17 x 512 2.046 BF
inputs = _conv2d_fixed_padding(inputs, 512, 3)
#51 conv 18 1 x 1/ 1 17 x 17 x 512 -> 17 x 17 x 18 0.005 BF
inputs = _conv2d_fixed_padding(inputs, 18, 1)
#52 yolo
detect_3 = _detection_layer(inputs, num_classes, _ANCHORS[6:9], img_size, data_format)
detect_3 = tf.identity(detect_3, name='detect_3')
detections = tf.concat([detect_1, detect_2, detect_3], axis=1)
detections = tf.identity(detections, name='detections')
return detections
def yolo_v3_tiny(inputs,
num_classes,
is_training=False,
data_format='NCHW',
reuse=False):
"""
Creates YOLO v3 tiny model.
:param inputs: a 4-D tensor of size [batch_size, height, width, channels].
Dimension batch_size may be undefined. The channel order is RGB.
:param num_classes: number of predicted classes.
:param is_training: whether is training or not.
:param data_format: data format NCHW or NHWC.
:param reuse: whether or not the network and its variables should be reused.
:return:
"""
# it will be needed later on
img_size = inputs.get_shape().as_list()[1:3]
# transpose the inputs to NCHW
if data_format == 'NCHW':
inputs = tf.transpose(inputs, [0, 3, 1, 2])
# normalize values to range [0..1]
inputs = inputs / 255
# set batch norm params
batch_norm_params = {
'decay': _BATCH_NORM_DECAY,
'epsilon': _BATCH_NORM_EPSILON,
'scale': True,
'is_training': is_training,
'fused': None, # Use fused batch norm if possible.
}
# Set activation_fn and parameters for conv2d, batch_norm.
with slim.arg_scope(
[slim.conv2d, slim.batch_norm, _fixed_padding, slim.max_pool2d],
data_format=data_format):
with slim.arg_scope([slim.conv2d, slim.batch_norm, _fixed_padding],
reuse=reuse):
with slim.arg_scope([slim.conv2d],
normalizer_fn=slim.batch_norm,
normalizer_params=batch_norm_params,
biases_initializer=None,
activation_fn=lambda x: tf.nn.leaky_relu(
x, alpha=_LEAKY_RELU)):
with tf.variable_scope('yolo-v3-tiny'):
#CSPDARKENT BEGIN
net = _conv2d_fixed_padding(inputs, 32, kernel_size=3)
net = _conv2d_fixed_padding(inputs, 64, kernel_size=3)
net, _ = _tiny_res_block(net, 64, data_format)
net, _ = _tiny_res_block(net, 128, data_format)
net, feat = _tiny_res_block(net, 256, data_format)
net = _conv2d_fixed_padding(net, 512, kernel_size=3)
feat2 = net
#CSPDARKNET END
net = _conv2d_fixed_padding(feat2, 256, kernel_size=1)
route = net
net = _conv2d_fixed_padding(route, 512, kernel_size=3)
detect_1 = _detection_layer(net, num_classes,
_ANCHORS[3:6], img_size,
data_format)
detect_1 = tf.identity(detect_1, name='detect_1')
net = _conv2d_fixed_padding(route, 128, kernel_size=1)
upsample_size = feat.get_shape().as_list()
net = _upsample(net, upsample_size, data_format)
net = tf.concat([feat, net],
axis=1 if data_format == 'NCHW' else 3)
net = _conv2d_fixed_padding(net, 256, kernel_size=3)
detect_2 = _detection_layer(net, num_classes,
_ANCHORS[1:4], img_size,
data_format)
detect_2 = tf.identity(detect_2, name='detect_2')
detections = tf.concat([detect_1, detect_2], axis=1)
detections = tf.identity(detections, name='detections')
return detections