def multibox_head(source_layers, num_priors, normalizations=None, softmax=True):
num_classes = 2
class_activation = 'softmax' if softmax else 'sigmoid'
mbox_conf = []
mbox_loc = []
mbox_quad = []
mbox_rbox = []
for i in range(len(source_layers)):
x = source_layers[i]
name = x.name.split('/')[0]
# normalize
if normalizations is not None and normalizations[i] > 0:
name = name + '_norm'
x = Normalize(normalizations[i], name=name)(x)
# confidence
name1 = name + '_mbox_conf'
x1 = Conv2D(num_priors[i] * num_classes, (3, 5), padding='same', name=name1)(x)
x1 = Flatten(name=name1+'_flat')(x1)
mbox_conf.append(x1)
# location, Delta(x,y,w,h)
name2 = name + '_mbox_loc'
x2 = Conv2D(num_priors[i] * 4, (3, 5), padding='same', name=name2)(x)
x2 = Flatten(name=name2+'_flat')(x2)
mbox_loc.append(x2)
# quadrilateral, Delta(x1,y1,x2,y2,x3,y3,x4,y4)
name3 = name + '_mbox_quad'
x3 = Conv2D(num_priors[i] * 8, (3, 5), padding='same', name=name3)(x)
x3 = Flatten(name=name3+'_flat')(x3)
mbox_quad.append(x3)
# rotated rectangle, Delta(x1,y1,x2,y2,h)
name4 = name + '_mbox_rbox'
x4 = Conv2D(num_priors[i] * 5, (3, 5), padding='same', name=name4)(x)
x4 = Flatten(name=name4+'_flat')(x4)
mbox_rbox.append(x4)
mbox_conf = concatenate(mbox_conf, axis=1, name='mbox_conf')
mbox_conf = Reshape((-1, num_classes), name='mbox_conf_logits')(mbox_conf)
mbox_conf = Activation(class_activation, name='mbox_conf_final')(mbox_conf)
mbox_loc = concatenate(mbox_loc, axis=1, name='mbox_loc')
mbox_loc = Reshape((-1, 4), name='mbox_loc_final')(mbox_loc)
mbox_quad = concatenate(mbox_quad, axis=1, name='mbox_quad')
mbox_quad = Reshape((-1, 8), name='mbox_quad_final')(mbox_quad)
mbox_rbox = concatenate(mbox_rbox, axis=1, name='mbox_rbox')
mbox_rbox = Reshape((-1, 5), name='mbox_rbox_final')(mbox_rbox)
predictions = concatenate([mbox_loc, mbox_quad, mbox_rbox, mbox_conf], axis=2, name='predictions')
return predictions
def multibox_head(source_layers,
num_priors,
num_classes,
normalizations=None,
softmax=True):
mbox_conf = []
mbox_loc = []
for i in range(len(source_layers)):
x = source_layers[i]
name = x.name.split('/')[0]
# normalize
if normalizations is not None and normalizations[i] > 0:
name = name + '_norm'
x = Normalize(normalizations[i], name=name)(x)
# confidence
name1 = name + '_mbox_conf'
x1 = Conv2D(num_priors[i] * num_classes, (1, 5),
padding='same',
name=name1)(x)
x1 = Flatten(name=name1 + '_flat')(x1)
mbox_conf.append(x1)
# location
name2 = name + '_mbox_loc'
x2 = Conv2D(num_priors[i] * 4, (1, 5), padding='same', name=name2)(x)
x2 = Flatten(name=name2 + '_flat')(x2)
mbox_loc.append(x2)
mbox_loc = concatenate(mbox_loc, axis=1, name='mbox_loc')
mbox_loc = Reshape((-1, 4), name='mbox_loc_final')(mbox_loc)
mbox_conf = concatenate(mbox_conf, axis=1, name='mbox_conf')
mbox_conf = Reshape((-1, num_classes), name='mbox_conf_logits')(mbox_conf)
if softmax:
mbox_conf = Activation('softmax', name='mbox_conf_final')(mbox_conf)
else:
mbox_conf = Activation('sigmoid', name='mbox_conf_final')(mbox_conf)
predictions = concatenate([mbox_loc, mbox_conf],
axis=2,
name='predictions')
return predictions
def multibox_head(source_layers, num_priors, normalizations=None, softmax=True):
num_classes = 2
class_activation = 'softmax' if softmax else 'sigmoid'
mbox_conf = []
mbox_loc = []
link_interlayer_conf = []
link_crosslayer_conf = []
for i in range(len(source_layers)):
x = source_layers[i]
name = x.name.split('/')[0]
# normalize
if normalizations is not None and normalizations[i] > 0:
name = name + '_norm'
x = Normalize(normalizations[i], name=name)(x)
# confidence
name1 = name + '_mbox_conf'
x1 = Conv2D(num_priors[i] * num_classes, 3, padding='same', name=name1)(x)
x1 = Flatten(name=name1+'_flat')(x1)
mbox_conf.append(x1)
# location
name2 = name + '_mbox_loc'
x2 = Conv2D(num_priors[i] * 5, 3, padding='same', name=name2)(x)
x2 = Flatten(name=name2+'_flat')(x2)
mbox_loc.append(x2)
# link interlayer confidenc
name3 = name + '_link_interlayer_conf'
x3 = Conv2D(num_priors[i] * num_classes * 8, 3, padding='same', name=name3)(x)
x3 = Flatten(name=name3+'_flat')(x3)
link_interlayer_conf.append(x3)
# link crosslayer confidenc
name4 = name + '_link_crosslayer_conf'
x4 = Conv2D(num_priors[i] * num_classes * 4, 3, padding='same', name=name4)(x)
x4 = Flatten(name=name4+'_flat')(x4)
link_crosslayer_conf.append(x4)
mbox_conf = concatenate(mbox_conf, axis=1, name='mbox_conf')
mbox_conf = Reshape((-1, num_classes), name='mbox_conf_logits')(mbox_conf)
mbox_conf = Activation(class_activation, name='mbox_conf_final')(mbox_conf)
mbox_loc = concatenate(mbox_loc, axis=1, name='mbox_loc')
mbox_loc = Reshape((-1, 5), name='mbox_loc_final')(mbox_loc)
#link_interlayer_conf = concatenate(link_interlayer_conf, axis=1, name='link_interlayer_conf')
#link_interlayer_conf = Reshape((-1, num_classes * 8), name='link_interlayer_conf_logits')(link_interlayer_conf)
#link_interlayer_conf = Activation(class_activation, name='link_interlayer_conf_final')(link_interlayer_conf)
#link_crosslayer_conf = concatenate(link_crosslayer_conf, axis=1, name='link_crosslayer_conf')
#link_crosslayer_conf = Reshape((-1, num_classes * 4), name='link_crosslayer_conf_logits')(link_crosslayer_conf)
#link_crosslayer_conf = Activation(class_activation, name='link_crosslayer_conf_final')(link_crosslayer_conf)
link_interlayer_conf = concatenate(link_interlayer_conf, axis=1, name='link_interlayer_conf')
link_interlayer_conf = Reshape((-1, num_classes), name='link_interlayer_conf_logits')(link_interlayer_conf)
link_interlayer_conf = Activation(class_activation, name='link_interlayer_conf_softmax')(link_interlayer_conf)
link_interlayer_conf = Reshape((-1, num_classes * 8), name='link_interlayer_conf_final')(link_interlayer_conf)
link_crosslayer_conf = concatenate(link_crosslayer_conf, axis=1, name='link_crosslayer_conf')
link_crosslayer_conf = Reshape((-1, num_classes), name='link_crosslayer_conf_logits')(link_crosslayer_conf)
link_crosslayer_conf = Activation(class_activation, name='link_crosslayer_conf_softmax')(link_crosslayer_conf)
link_crosslayer_conf = Reshape((-1, num_classes * 4), name='link_crosslayer_conf_final')(link_crosslayer_conf)
predictions = concatenate([
mbox_conf,
mbox_loc,
link_interlayer_conf,
link_crosslayer_conf
], axis=2, name='predictions')
return predictions
def multibox_head(source_layers,
num_priors,
normalizations=None,
softmax=True):
num_classes = 6
class_activation = 'softmax' if softmax else 'sigmoid'
mbox_conf = []
mbox_loc = []
mbox_quad = []
mbox_rbox = []
for i in range(len(source_layers)):
x = source_layers[
i] #i=0 => image size of x : (64,64), shape of x : [None, 64, 64, 512]
name = x.name.split('/')[0]
# normalize
if normalizations is not None and normalizations[i] > 0:
name = name + '_norm'
x = Normalize(normalizations[i], name=name)(x)
# confidence
name1 = name + '_mbox_conf'
x1 = Conv2D(
num_priors[i] * num_classes, (3, 5), padding='same', name=name1
)(
x
) #num_priors[i]=14, num_classes = 6 => x1 shape : [None, 64, 64, 14*6]
x1 = Flatten(name=name1 + '_flat')(
x1) # shape of x1: [batch_size, 64*64*14*6], per class : 64*64*14
mbox_conf.append(x1)
# location, Delta(x,y,w,h)
name2 = name + '_mbox_loc'
x2 = Conv2D(num_priors[i] * 4, (3, 5), padding='same', name=name2)(x)
x2 = Flatten(name=name2 + '_flat')(x2)
mbox_loc.append(x2) # shape of x2: [batch_size, 64*64*14*4]
# quadrilateral, Delta(x1,y1,x2,y2,x3,y3,x4,y4)
name3 = name + '_mbox_quad'
x3 = Conv2D(num_priors[i] * 8, (3, 5), padding='same', name=name3)(x)
x3 = Flatten(name=name3 + '_flat')(x3)
mbox_quad.append(x3) # shape of x3: [batch_size, 64*64*14*8]
# rotated rectangle, Delta(x1,y1,x2,y2,h)
name4 = name + '_mbox_rbox'
x4 = Conv2D(num_priors[i] * 5, (3, 5), padding='same', name=name4)(x)
x4 = Flatten(name=name4 + '_flat')(x4)
mbox_rbox.append(x4) # shape of x4: [batch_size, 64*64*14*5]
mbox_conf = concatenate(mbox_conf, axis=1, name='mbox_conf')
mbox_conf = Reshape(
(-1, num_classes),
name='mbox_conf_logits')(mbox_conf) #shape : [batch_size, 64*64*14, 6]
mbox_conf = Activation(class_activation, name='mbox_conf_final')(mbox_conf)
mbox_loc = concatenate(mbox_loc, axis=1, name='mbox_loc')
mbox_loc = Reshape(
(-1, 4),
name='mbox_loc_final')(mbox_loc) #shape : [batch_size, 64*64*14, 4]
mbox_quad = concatenate(mbox_quad, axis=1, name='mbox_quad')
mbox_quad = Reshape(
(-1, 8),
name='mbox_quad_final')(mbox_quad) #shape : [batch_size, 64*64*14, 8]
mbox_rbox = concatenate(mbox_rbox, axis=1, name='mbox_rbox')
mbox_rbox = Reshape(
(-1, 5),
name='mbox_rbox_final')(mbox_rbox) #shape : [batch_size, 64*64*14, 5]
predictions = concatenate([mbox_loc, mbox_quad, mbox_rbox, mbox_conf],
axis=2,
name='predictions')
return predictions # shape : (batch_size,Nums_priors,4+8+5+6) = (batch_size, 76454, 23)