def resnet_101_base(img_batch, is_training):
'''
this code is from light-head rcnn.
It is convenient to freeze blocks.
@author: jemmy li
@contact: [email protected]
'''
bottleneck = resnet_v1.bottleneck
blocks = [
modified_resnet_v1_block('block1', bottleneck,
[(256, 64, 1, 1)] * 2 + [(256, 64, 1, 1)]),
modified_resnet_v1_block('block2', bottleneck,
[(512, 128, 2, 1)] + [(512, 128, 1, 1)] * 3),
modified_resnet_v1_block('block3', bottleneck, [(1024, 256, 2, 1)] +
[(1024, 256, 1, 1)] * 22)
]
with slim.arg_scope(
resnet_arg_scope(is_training=False)): # freeze the base conv_net
with tf.variable_scope('resnet_v1_101', 'resnet_v1_101'):
net = resnet_utils.conv2d_same(img_batch,
64,
7,
stride=2,
scope='conv1')
net = slim.max_pool2d(net, [3, 3],
stride=2,
padding='SAME',
scope='pool1')
not_freezed = [False
] * cfgs.FIXED_BLOCKS + (4 - cfgs.FIXED_BLOCKS) * [True]
# Fixed_Blocks can be 1~3
with slim.arg_scope(
resnet_arg_scope(is_training=is_training and not_freezed[0])):
C2, _ = resnet_v1.resnet_v1(net,
blocks[0:1],
global_pool=False,
include_root_block=False,
scope='resnet_v1_101')
with slim.arg_scope(
resnet_arg_scope(is_training=is_training and not_freezed[1])):
C3, _ = resnet_v1.resnet_v1(C2,
blocks[1:2],
global_pool=False,
include_root_block=False,
scope='resnet_v1_101')
with slim.arg_scope(
resnet_arg_scope(is_training=is_training and not_freezed[2])):
C4, _ = resnet_v1.resnet_v1(C3,
blocks[2:3],
global_pool=False,
include_root_block=False,
scope='resnet_v1_101')
return C4
def restnet_head(input, is_training, scope):
block4 = [
modified_resnet_v1_block('block4', resnet_v1.bottleneck,
[(2048, 512, 1, 2)] + [(2048, 512, 1, 2)] * 2)
]
with slim.arg_scope(resnet_arg_scope(is_training=is_training)):
C5_flatten, _ = resnet_v1.resnet_v1(input,
block4,
global_pool=True,
spatial_squeeze=True,
include_root_block=False,
scope=scope)
# global average pooling C5 to obtain fc layers
return C5_flatten