def build_net(self, Iter):
im, bboxes, nums = Iter.get_next()
im.set_shape(
tf.TensorShape(
[None, self.config.img_size, self.config.img_size, 3]))
im = im / 255 * 2 - 1
batch_m = tf.shape(im)[0]
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(
is_training=self.config.is_train,
batch_norm_decay=0.9997,
stddev=0.03)):
logits, end_points = mobilenet_v1.mobilenet_v1(
im, min_depth=16, is_training=self.config.is_train)
var_pre = tf.global_variables()[1:]
net11 = end_points['Conv2d_11_pointwise']
net13 = end_points['Conv2d_13_pointwise']
net14 = new_conv2d(net13, 512, 2, 'conv14')
net15 = new_conv2d(net14, 256, 2, 'conv15')
net16 = new_conv2d(net15, 256, 2, 'conv16')
net17 = new_conv2d(net16, 128, 2, 'conv17')
weights_init = tf.truncated_normal_initializer(stddev=0.03)
net = [net11, net13, net14, net15, net16, net17]
with slim.arg_scope([slim.conv2d],
activation_fn=None,
weights_initializer=weights_init,
weights_regularizer=slim.l2_regularizer(
self.config.weight_decay),
normalizer_fn=None):
for i in range(len(self.config.num_anchors)):
with tf.variable_scope('BoxPredictor_%d' % i):
with tf.variable_scope('ClassPredictor'):
net_cls = slim.conv2d(
net[i], self.config.num_anchors[i] *
(self.config.num_cls + 1), [1, 1])
with tf.variable_scope('BoxEncodingPredictor'):
net_box = slim.conv2d(
net[i],
self.config.num_anchors[i] * 4,
[1, 1],
)
net_cls = tf.reshape(
net_cls,
(batch_m, -1, self.config.num_cls + 1),
)
net_box = tf.reshape(
net_box,
(batch_m, -1, 4),
)
net[i] = tf.concat([net_cls, net_box], axis=-1)
net = tf.concat(net, axis=1, name='net')
loss, Num = self.get_loss(net, bboxes, nums)
return tf.reduce_sum(loss) / tf.reduce_sum(Num + 1e-10), var_pre
def build_net(self, ):
self.im_input = tf.placeholder(tf.string, name='input')
im, b_scale, = self.handle_im(self.im_input, size=self.config.img_size)
im.set_shape(tf.TensorShape([None, self.config.img_size, self.config.img_size, 3]))
im = im / 255 * 2 - 1
batch_m = tf.shape(im)[0]
with slim.arg_scope(mobilenet_v1.mobilenet_v1_arg_scope(is_training=self.config.is_train, batch_norm_decay=0.9)):
logits, end_points = mobilenet_v1.mobilenet_v1(im, min_depth=16, is_training=self.config.is_train)
var_pre = tf.global_variables()[1:]
net11 = end_points['Conv2d_11_pointwise']
net13 = end_points['Conv2d_13_pointwise']
net14 = new_conv2d(net13, 512, 2, 'conv14')
net15 = new_conv2d(net14, 256, 2, 'conv15')
net16 = new_conv2d(net15, 256, 2, 'conv16')
net17 = new_conv2d(net16, 128, 2, 'conv17')
weights_init = tf.truncated_normal_initializer(stddev=0.03)
net = [net11, net13, net14, net15, net16, net17]
with slim.arg_scope([slim.conv2d], activation_fn=None, weights_initializer=weights_init,
weights_regularizer=slim.l2_regularizer(self.config.weight_decay), normalizer_fn=None):
for i in range(len(self.config.num_anchors)):
with tf.variable_scope('BoxPredictor_%d' % i):
with tf.variable_scope('ClassPredictor'):
net_cls = slim.conv2d(net[i], self.config.num_anchors[i] * (self.config.num_cls + 1), [1, 1])
with tf.variable_scope('BoxEncodingPredictor'):
net_box = slim.conv2d(net[i], self.config.num_anchors[i] * 4, [1, 1], )
net_cls = tf.reshape(net_cls, (batch_m, -1, self.config.num_cls + 1), )
net_box = tf.reshape(net_box, (batch_m, -1, 4), )
net[i] = tf.concat([net_cls, net_box], axis=-1)
net = tf.concat(net, axis=1, name='net')
print(net)
pre_bboxes = loc2bbox(net[..., -4:] * tf.constant([0.1, 0.1, 0.2, 0.2], dtype=tf.float32), self.anchors)
pre = predict(pre_bboxes[0], net[0][..., :-4], size=self.config.img_size, c_thresh=1e-2,num_cls=self.config.num_cls)
self.result = tf.concat([pre[..., :4] * b_scale, pre[..., 4:]], axis=-1, name='out')
print(self.result)
def build_net(self, ):
self.im_input = tf.placeholder(tf.string, name='input')
im, b_scale, = self.handle_im(self.im_input, size=self.config.img_size)
im.set_shape(
tf.TensorShape(
[None, self.config.img_size, self.config.img_size, 3]))
im = im / 255 * 2 - 1
batch_m = tf.shape(im)[0]
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(
is_training=self.config.is_train, batch_norm_decay=0.9)):
logits, end_points = mobilenet_v1.mobilenet_v1(
im, min_depth=16, is_training=self.config.is_train)
var_pre = tf.global_variables()[1:]
net11 = end_points['Conv2d_11_pointwise']
net13 = end_points['Conv2d_13_pointwise']
net14 = new_conv2d(net13, 512, 2, 'conv14')
net15 = new_conv2d(net14, 256, 2, 'conv15')
net16 = new_conv2d(net15, 256, 2, 'conv16')
net17 = new_conv2d(net16, 128, 2, 'conv17')
with tf.variable_scope('detect_cls'):
with slim.arg_scope([slim.conv2d],
activation_fn=None,
weights_regularizer=slim.l2_regularizer(
self.config.weight_decay)):
net11_cls = slim.conv2d(net11,
self.config.num_anchors[0] *
(self.config.num_cls + 1), [1, 1],
scope='net11')
net13_cls = slim.conv2d(net13,
self.config.num_anchors[1] *
(self.config.num_cls + 1), [1, 1],
scope='net13')
net14_cls = slim.conv2d(net14,
self.config.num_anchors[2] *
(self.config.num_cls + 1), [1, 1],
scope='net14')
net15_cls = slim.conv2d(net15,
self.config.num_anchors[3] *
(self.config.num_cls + 1), [1, 1],
scope='net15')
net16_cls = slim.conv2d(net16,
self.config.num_anchors[4] *
(self.config.num_cls + 1), [1, 1],
scope='net16')
net17_cls = slim.conv2d(net17,
self.config.num_anchors[5] *
(self.config.num_cls + 1), [1, 1],
scope='net17')
with tf.variable_scope('detect_box'):
with slim.arg_scope([slim.conv2d],
activation_fn=None,
weights_regularizer=slim.l2_regularizer(
self.config.weight_decay)):
net11_box = slim.conv2d(net11,
self.config.num_anchors[0] * 4, [1, 1],
scope='net11')
net13_box = slim.conv2d(net13,
self.config.num_anchors[1] * 4, [1, 1],
scope='net13')
net14_box = slim.conv2d(net14,
self.config.num_anchors[2] * 4, [1, 1],
scope='net14')
net15_box = slim.conv2d(net15,
self.config.num_anchors[3] * 4, [1, 1],
scope='net15')
net16_box = slim.conv2d(net16,
self.config.num_anchors[4] * 4, [1, 1],
scope='net16')
net17_box = slim.conv2d(net17,
self.config.num_anchors[5] * 4, [1, 1],
scope='net17')
net11 = tf.concat([net11_cls, net11_box], axis=-1)
net13 = tf.concat([net13_cls, net13_box], axis=-1)
net14 = tf.concat([net14_cls, net14_box], axis=-1)
net15 = tf.concat([net15_cls, net15_box], axis=-1)
net16 = tf.concat([net16_cls, net16_box], axis=-1)
net17 = tf.concat([net17_cls, net17_box], axis=-1)
net11 = tf.reshape(
net11,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net13 = tf.reshape(
net13,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net14 = tf.reshape(
net14,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net15 = tf.reshape(
net15,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net16 = tf.reshape(
net16,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net17 = tf.reshape(
net17,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net = tf.concat((net11, net13, net14, net15, net16, net17),
axis=1,
name='net')
print(net)
pre_bboxes = loc2bbox(
net[..., -4:] *
tf.constant([0.1, 0.1, 0.2, 0.2], dtype=tf.float32), self.anchors)
pre = predict(pre_bboxes[0],
net[0][..., :-4],
size=self.config.img_size,
c_thresh=1e-2,
num_cls=self.config.num_cls)
self.result = tf.concat([pre[..., :4] * b_scale, pre[..., 4:]],
axis=-1,
name='out')
print(self.result)
def build_net(self, Iter):
im, bboxes, nums = Iter.get_next()
im.set_shape(
tf.TensorShape(
[None, self.config.img_size, self.config.img_size, 3]))
im = im / 255 * 2 - 1
batch_m = tf.shape(im)[0]
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(
is_training=self.config.is_train,
batch_norm_decay=0.9997,
stddev=0.03)):
logits, end_points = mobilenet_v1.mobilenet_v1(
im, min_depth=16, is_training=self.config.is_train)
var_pre = tf.global_variables()[1:]
net11 = end_points['Conv2d_11_pointwise']
net13 = end_points['Conv2d_13_pointwise']
net14 = new_conv2d(net13, 512, 2, 'conv14')
net15 = new_conv2d(net14, 256, 2, 'conv15')
net16 = new_conv2d(net15, 256, 2, 'conv16')
net17 = new_conv2d(net16, 128, 2, 'conv17')
weights_init = tf.truncated_normal_initializer(stddev=0.03)
with tf.variable_scope('detect_cls'):
with slim.arg_scope([slim.conv2d],
activation_fn=None,
weights_initializer=weights_init,
weights_regularizer=slim.l2_regularizer(
self.config.weight_decay)):
net11_cls = slim.conv2d(net11,
self.config.num_anchors[0] *
(self.config.num_cls + 1), [1, 1],
scope='net11')
net13_cls = slim.conv2d(net13,
self.config.num_anchors[1] *
(self.config.num_cls + 1), [1, 1],
scope='net13')
net14_cls = slim.conv2d(net14,
self.config.num_anchors[2] *
(self.config.num_cls + 1), [1, 1],
scope='net14')
net15_cls = slim.conv2d(net15,
self.config.num_anchors[3] *
(self.config.num_cls + 1), [1, 1],
scope='net15')
net16_cls = slim.conv2d(net16,
self.config.num_anchors[4] *
(self.config.num_cls + 1), [1, 1],
scope='net16')
net17_cls = slim.conv2d(net17,
self.config.num_anchors[5] *
(self.config.num_cls + 1), [1, 1],
scope='net17')
with tf.variable_scope('detect_box'):
with slim.arg_scope([slim.conv2d],
activation_fn=None,
weights_initializer=weights_init,
weights_regularizer=slim.l2_regularizer(
self.config.weight_decay)):
net11_box = slim.conv2d(net11,
self.config.num_anchors[0] * 4, [1, 1],
scope='net11')
net13_box = slim.conv2d(net13,
self.config.num_anchors[1] * 4, [1, 1],
scope='net13')
net14_box = slim.conv2d(net14,
self.config.num_anchors[2] * 4, [1, 1],
scope='net14')
net15_box = slim.conv2d(net15,
self.config.num_anchors[3] * 4, [1, 1],
scope='net15')
net16_box = slim.conv2d(net16,
self.config.num_anchors[4] * 4, [1, 1],
scope='net16')
net17_box = slim.conv2d(net17,
self.config.num_anchors[5] * 4, [1, 1],
scope='net17')
net11 = tf.concat([net11_cls, net11_box], axis=-1)
net13 = tf.concat([net13_cls, net13_box], axis=-1)
net14 = tf.concat([net14_cls, net14_box], axis=-1)
net15 = tf.concat([net15_cls, net15_box], axis=-1)
net16 = tf.concat([net16_cls, net16_box], axis=-1)
net17 = tf.concat([net17_cls, net17_box], axis=-1)
net11 = tf.reshape(
net11,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net13 = tf.reshape(
net13,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net14 = tf.reshape(
net14,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net15 = tf.reshape(
net15,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net16 = tf.reshape(
net16,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net17 = tf.reshape(
net17,
(batch_m, -1, self.config.num_cls + 1 + 4),
)
net = tf.concat((net11, net13, net14, net15, net16, net17),
axis=1,
name='net')
loss, Num = self.get_loss(net, bboxes, nums)
return tf.reduce_sum(loss) / tf.reduce_sum(Num + 1e-10), var_pre