def inference(images, num_classes, scope):
with tf.op_scope([images], scope):
with scopes.arg_scope([ops.conv2d, ops.fc, ops.dropout],
is_training=True):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(images,
num_filters_out=64,
kernel_size=[11, 11],
stride=4,
stddev=1e-1)
with tf.variable_scope('pool1'):
pool1 = ops.max_pool(conv1,
kernel_size=3,
stride=2,
padding='VALID')
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(pool1,
num_filters_out=192,
kernel_size=1,
stride=1,
stddev=1e-1)
with tf.variable_scope('pool2'):
pool2 = ops.max_pool(conv2,
kernel_size=3,
stride=2,
padding='VALID')
with tf.variable_scope('conv3'):
conv3 = ops.conv2d(pool2,
num_filters_out=384,
kernel_size=3,
stride=1,
stddev=1e-1)
with tf.variable_scope('conv4'):
conv4 = ops.conv2d(conv3,
num_filters_out=256,
kernel_size=3,
stride=1,
stddev=1e-1)
with tf.variable_scope('conv5'):
conv5 = ops.conv2d(conv4,
num_filters_out=256,
kernel_size=3,
stride=1,
stddev=1e-1)
with tf.variable_scope('pool3'):
pool3 = ops.max_pool(conv5,
kernel_size=3,
stride=2,
padding='VALID')
flattened = ops.flatten(pool3, scope='flatten')
with tf.variable_scope('fc1'):
fc1 = ops.fc(flattened, num_units_out=4096)
with tf.variable_scope('fc2'):
fc2 = ops.fc(fc1, num_units_out=4096)
with tf.variable_scope('fc3'):
fc3 = ops.fc(fc2, activation=None, num_units_out=num_classes)
return fc3
def _res_block(inputs, num_filters_start, half_input):
print (inputs.get_shape())
num_filters_in = inputs.get_shape()[-1]
num_filters_out = num_filters_start * 4
need_branch1 = (num_filters_in != num_filters_out)
# TODO tf does not support conv2d with stride > kernel_size
# Workaround: use a pooling to replace this downsampling
#stride1 = 2 if half_input else 1
stride1 = 1
if half_input:
with tf.variable_scope('downsample'):
inputs = ops.max_pool(
inputs,
kernel_size=3,
stride=2,
padding='SAME')
if need_branch1:
with tf.variable_scope('branch1'):
branch1 = ops.conv2d(
inputs,
num_filters_out=num_filters_out,
kernel_size=1,
stride=stride1)
with tf.variable_scope('branch2'):
with tf.variable_scope('a'):
branch2a = ops.conv2d(
inputs,
num_filters_out=num_filters_start,
kernel_size=1,
stride=stride1)
with tf.variable_scope('b'):
branch2b = ops.conv2d(
branch2a,
num_filters_out=num_filters_start,
kernel_size=3,
stride=1)
with tf.variable_scope('c'):
branch2c = ops.conv2d(
branch2b,
num_filters_out=num_filters_out,
kernel_size=1,
stride=1,
activation=None)
branch_sum = branch2c + branch1 if need_branch1 else branch2c
ret = tf.nn.relu(branch_sum)
return ret
def _res_block(inputs, num_filters_start, half_input):
print(inputs.get_shape())
num_filters_in = inputs.get_shape()[-1]
num_filters_out = num_filters_start * 4
need_branch1 = (num_filters_in != num_filters_out)
# TODO tf does not support conv2d with stride > kernel_size
# Workaround: use a pooling to replace this downsampling
#stride1 = 2 if half_input else 1
stride1 = 1
if half_input:
with tf.variable_scope('downsample'):
inputs = ops.max_pool(inputs,
kernel_size=3,
stride=2,
padding='SAME')
if need_branch1:
with tf.variable_scope('branch1'):
branch1 = ops.conv2d(inputs,
num_filters_out=num_filters_out,
kernel_size=1,
stride=stride1)
with tf.variable_scope('branch2'):
with tf.variable_scope('a'):
branch2a = ops.conv2d(inputs,
num_filters_out=num_filters_start,
kernel_size=1,
stride=stride1)
with tf.variable_scope('b'):
branch2b = ops.conv2d(branch2a,
num_filters_out=num_filters_start,
kernel_size=3,
stride=1)
with tf.variable_scope('c'):
branch2c = ops.conv2d(branch2b,
num_filters_out=num_filters_out,
kernel_size=1,
stride=1,
activation=None)
branch_sum = branch2c + branch1 if need_branch1 else branch2c
ret = tf.nn.relu(branch_sum)
return ret
def inference(inputs, num_classes, scope):
batch_norm_params = {
# Decay for the batch_norm moving averages.
'decay': 0.9997,
# epsilon to prevent 0s in variance.
'epsilon': 0.001,
}
with tf.op_scope([inputs], scope, 'resnet50'):
# TODO how to implement scale layer?
with scopes.arg_scope([ops.conv2d],
batch_norm_params=batch_norm_params):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(inputs,
num_filters_out=64,
kernel_size=7,
stride=2)
with tf.variable_scope('poo1'):
pool1 = ops.max_pool(conv1,
kernel_size=3,
stride=2,
padding='SAME')
with tf.variable_scope('res2'):
res2 = _res_group(pool1,
num_filters_start=64,
num_blocks=3,
first_group=True)
with tf.variable_scope('res3'):
res3 = _res_group(res2, num_filters_start=128, num_blocks=8)
with tf.variable_scope('res4'):
res4 = _res_group(res3, num_filters_start=256, num_blocks=36)
with tf.variable_scope('res5'):
res5 = _res_group(res4, num_filters_start=512, num_blocks=3)
with tf.variable_scope('pool5'):
pool5 = ops.avg_pool(res5,
kernel_size=7,
stride=1,
padding='SAME')
flattened = ops.flatten(pool5, scope='flatten')
with tf.variable_scope('fc'):
fc = ops.fc(flattened,
activation=None,
num_units_out=num_classes)
return fc
def inference(inputs, num_classes, scope):
batch_norm_params = {
# Decay for the batch_norm moving averages.
'decay': 0.9997,
# epsilon to prevent 0s in variance.
'epsilon': 0.001,
}
with tf.op_scope([inputs], scope, 'resnet50'):
# TODO how to implement scale layer?
with scopes.arg_scope([ops.conv2d], batch_norm_params=batch_norm_params):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(
inputs,
num_filters_out=64,
kernel_size=7,
stride=2)
with tf.variable_scope('poo1'):
pool1 = ops.max_pool(
conv1,
kernel_size=3,
stride=2,
padding='SAME')
with tf.variable_scope('res2'):
res2 = _res_group(pool1, num_filters_start=64, num_blocks=3, first_group=True)
with tf.variable_scope('res3'):
res3 = _res_group(res2, num_filters_start=128, num_blocks=4)
with tf.variable_scope('res4'):
res4 = _res_group(res3, num_filters_start=256, num_blocks=23)
with tf.variable_scope('res5'):
res5 = _res_group(res4, num_filters_start=512, num_blocks=3)
with tf.variable_scope('pool5'):
pool5 = ops.avg_pool(
res5,
kernel_size=7,
stride=1,
padding='SAME')
flattened = ops.flatten(pool5, scope='flatten')
with tf.variable_scope('fc'):
fc = ops.fc(
flattened,
activation=None,
num_units_out=num_classes)
return fc
def inference(inputs, num_classes, scope):
with tf.op_scope([inputs], scope, 'vgg19'):
with scopes.arg_scope([ops.conv2d, ops.fc, ops.batch_norm, ops.dropout], is_training=True):
# block 1
with tf.variable_scope('block1'):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(
inputs,
num_filters_out=64,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(
conv1,
num_filters_out=64,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('pool'):
pool = ops.max_pool(
conv2,
kernel_size=2,
stride=2,
padding='VALID')
# block 2
with tf.variable_scope('block2'):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(
pool,
num_filters_out=128,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(
conv1,
num_filters_out=128,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('pool'):
pool = ops.max_pool(
conv2,
kernel_size=2,
stride=2,
padding='VALID')
# block 3
with tf.variable_scope('block3'):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(
pool,
num_filters_out=256,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(
conv1,
num_filters_out=256,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv3'):
conv3 = ops.conv2d(
conv2,
num_filters_out=256,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('pool'):
pool = ops.max_pool(
conv3,
kernel_size=2,
stride=2,
padding='VALID')
# block 4
with tf.variable_scope('block4'):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(
pool,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(
conv1,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv3'):
conv3 = ops.conv2d(
conv2,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('pool'):
pool = ops.max_pool(
conv3,
kernel_size=2,
stride=2,
padding='VALID')
# block 5
with tf.variable_scope('block5'):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(
pool,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(
conv1,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv3'):
conv3 = ops.conv2d(
conv2,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('pool'):
pool = ops.max_pool(
conv3,
kernel_size=2,
stride=2,
padding='VALID')
flattened = ops.flatten(pool, scope='flatten')
# fc
with tf.variable_scope('fc1'):
fc1 = ops.fc(
flattened,
num_units_out=4096)
with tf.variable_scope('fc2'):
fc2 = ops.fc(
fc1,
num_units_out=4096)
with tf.variable_scope('fc3'):
fc3 = ops.fc(
fc2,
activation=None,
num_units_out=num_classes)
return fc3
def inference(inputs,
dropout_keep_prob=0.8,
num_classes=1000,
is_training=True,
restore_logits=True,
scope=''):
"""Latest Inception from http://arxiv.org/abs/1512.00567.
"Rethinking the Inception Architecture for Computer Vision"
Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens,
Zbigniew Wojna
Args:
inputs: a tensor of size [batch_size, height, width, channels].
dropout_keep_prob: dropout keep_prob.
num_classes: number of predicted classes.
is_training: whether is training or not.
restore_logits: whether or not the logits layers should be restored.
Useful for fine-tuning a model with different num_classes.
scope: Optional scope for op_scope.
Returns:
a list containing 'logits', 'aux_logits' Tensors.
"""
# end_points will collect relevant activations for external use, for example
# summaries or losses.
end_points = {}
batch_norm_params = {
# Decay for the batch_norm moving averages.
'decay': 0.9997,
# epsilon to prevent 0s in variance.
'epsilon': 0.001,
}
with tf.op_scope([inputs], scope, 'inception_v3'):
with scopes.arg_scope([ops.conv2d],
batch_norm_params=batch_norm_params):
with scopes.arg_scope([ops.conv2d, ops.max_pool, ops.avg_pool],
stride=1,
padding='VALID'):
# 299 x 299 x 3
end_points['conv0'] = ops.conv2d(inputs,
32, [3, 3],
stride=2,
scope='conv0')
# 149 x 149 x 32
end_points['conv1'] = ops.conv2d(end_points['conv0'],
32, [3, 3],
scope='conv1')
# 147 x 147 x 32
end_points['conv2'] = ops.conv2d(end_points['conv1'],
64, [3, 3],
padding='SAME',
scope='conv2')
# 147 x 147 x 64
end_points['pool1'] = ops.max_pool(end_points['conv2'], [3, 3],
stride=2,
scope='pool1')
# 73 x 73 x 64
end_points['conv3'] = ops.conv2d(end_points['pool1'],
80, [1, 1],
scope='conv3')
# 73 x 73 x 80.
end_points['conv4'] = ops.conv2d(end_points['conv3'],
192, [3, 3],
scope='conv4')
# 71 x 71 x 192.
end_points['pool2'] = ops.max_pool(end_points['conv4'], [3, 3],
stride=2,
scope='pool2')
# 35 x 35 x 192.
net = end_points['pool2']
# Inception blocks
with scopes.arg_scope([ops.conv2d, ops.max_pool, ops.avg_pool],
stride=1,
padding='SAME'):
# mixed: 35 x 35 x 256.
with tf.variable_scope('mixed_35x35x256a'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 64, [1, 1])
with tf.variable_scope('branch5x5'):
branch5x5 = ops.conv2d(net, 48, [1, 1])
branch5x5 = ops.conv2d(branch5x5, 64, [5, 5])
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 64, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 32, [1, 1])
net = tf.concat(
3, [branch1x1, branch5x5, branch3x3dbl, branch_pool])
end_points['mixed_35x35x256a'] = net
# mixed_1: 35 x 35 x 288.
with tf.variable_scope('mixed_35x35x288a'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 64, [1, 1])
with tf.variable_scope('branch5x5'):
branch5x5 = ops.conv2d(net, 48, [1, 1])
branch5x5 = ops.conv2d(branch5x5, 64, [5, 5])
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 64, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 64, [1, 1])
net = tf.concat(
3, [branch1x1, branch5x5, branch3x3dbl, branch_pool])
end_points['mixed_35x35x288a'] = net
# mixed_2: 35 x 35 x 288.
with tf.variable_scope('mixed_35x35x288b'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 64, [1, 1])
with tf.variable_scope('branch5x5'):
branch5x5 = ops.conv2d(net, 48, [1, 1])
branch5x5 = ops.conv2d(branch5x5, 64, [5, 5])
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 64, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 64, [1, 1])
net = tf.concat(
3, [branch1x1, branch5x5, branch3x3dbl, branch_pool])
end_points['mixed_35x35x288b'] = net
# mixed_3: 17 x 17 x 768.
with tf.variable_scope('mixed_17x17x768a'):
with tf.variable_scope('branch3x3'):
branch3x3 = ops.conv2d(net,
384, [3, 3],
stride=2,
padding='VALID')
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 64, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
branch3x3dbl = ops.conv2d(branch3x3dbl,
96, [3, 3],
stride=2,
padding='VALID')
with tf.variable_scope('branch_pool'):
branch_pool = ops.max_pool(net, [3, 3],
stride=2,
padding='VALID')
net = tf.concat(3, [branch3x3, branch3x3dbl, branch_pool])
end_points['mixed_17x17x768a'] = net
# mixed4: 17 x 17 x 768.
with tf.variable_scope('mixed_17x17x768b'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 192, [1, 1])
with tf.variable_scope('branch7x7'):
branch7x7 = ops.conv2d(net, 128, [1, 1])
branch7x7 = ops.conv2d(branch7x7, 128, [1, 7])
branch7x7 = ops.conv2d(branch7x7, 192, [7, 1])
with tf.variable_scope('branch7x7dbl'):
branch7x7dbl = ops.conv2d(net, 128, [1, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 128, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 128, [1, 7])
branch7x7dbl = ops.conv2d(branch7x7dbl, 128, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [1, 7])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(
3, [branch1x1, branch7x7, branch7x7dbl, branch_pool])
end_points['mixed_17x17x768b'] = net
# mixed_5: 17 x 17 x 768.
with tf.variable_scope('mixed_17x17x768c'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 192, [1, 1])
with tf.variable_scope('branch7x7'):
branch7x7 = ops.conv2d(net, 160, [1, 1])
branch7x7 = ops.conv2d(branch7x7, 160, [1, 7])
branch7x7 = ops.conv2d(branch7x7, 192, [7, 1])
with tf.variable_scope('branch7x7dbl'):
branch7x7dbl = ops.conv2d(net, 160, [1, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [1, 7])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [1, 7])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(
3, [branch1x1, branch7x7, branch7x7dbl, branch_pool])
end_points['mixed_17x17x768c'] = net
# mixed_6: 17 x 17 x 768.
with tf.variable_scope('mixed_17x17x768d'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 192, [1, 1])
with tf.variable_scope('branch7x7'):
branch7x7 = ops.conv2d(net, 160, [1, 1])
branch7x7 = ops.conv2d(branch7x7, 160, [1, 7])
branch7x7 = ops.conv2d(branch7x7, 192, [7, 1])
with tf.variable_scope('branch7x7dbl'):
branch7x7dbl = ops.conv2d(net, 160, [1, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [1, 7])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [1, 7])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(
3, [branch1x1, branch7x7, branch7x7dbl, branch_pool])
end_points['mixed_17x17x768d'] = net
# mixed_7: 17 x 17 x 768.
with tf.variable_scope('mixed_17x17x768e'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 192, [1, 1])
with tf.variable_scope('branch7x7'):
branch7x7 = ops.conv2d(net, 192, [1, 1])
branch7x7 = ops.conv2d(branch7x7, 192, [1, 7])
branch7x7 = ops.conv2d(branch7x7, 192, [7, 1])
with tf.variable_scope('branch7x7dbl'):
branch7x7dbl = ops.conv2d(net, 192, [1, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [1, 7])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [1, 7])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(
3, [branch1x1, branch7x7, branch7x7dbl, branch_pool])
end_points['mixed_17x17x768e'] = net
# Auxiliary Head logits
aux_logits = tf.identity(end_points['mixed_17x17x768e'])
with tf.variable_scope('aux_logits'):
aux_logits = ops.avg_pool(aux_logits, [5, 5],
stride=3,
padding='VALID')
aux_logits = ops.conv2d(aux_logits,
128, [1, 1],
scope='proj')
# Shape of feature map before the final layer.
shape = aux_logits.get_shape()
aux_logits = ops.conv2d(aux_logits,
768,
shape[1:3],
stddev=0.01,
padding='VALID')
aux_logits = ops.flatten(aux_logits)
aux_logits = ops.fc(aux_logits,
num_classes,
activation=None,
stddev=0.001,
restore=restore_logits)
end_points['aux_logits'] = aux_logits
# mixed_8: 8 x 8 x 1280.
# Note that the scope below is not changed to not void previous
# checkpoints.
# (TODO) Fix the scope when appropriate.
with tf.variable_scope('mixed_17x17x1280a'):
with tf.variable_scope('branch3x3'):
branch3x3 = ops.conv2d(net, 192, [1, 1])
branch3x3 = ops.conv2d(branch3x3,
320, [3, 3],
stride=2,
padding='VALID')
with tf.variable_scope('branch7x7x3'):
branch7x7x3 = ops.conv2d(net, 192, [1, 1])
branch7x7x3 = ops.conv2d(branch7x7x3, 192, [1, 7])
branch7x7x3 = ops.conv2d(branch7x7x3, 192, [7, 1])
branch7x7x3 = ops.conv2d(branch7x7x3,
192, [3, 3],
stride=2,
padding='VALID')
with tf.variable_scope('branch_pool'):
branch_pool = ops.max_pool(net, [3, 3],
stride=2,
padding='VALID')
net = tf.concat(3, [branch3x3, branch7x7x3, branch_pool])
end_points['mixed_17x17x1280a'] = net
# mixed_9: 8 x 8 x 2048.
with tf.variable_scope('mixed_8x8x2048a'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 320, [1, 1])
with tf.variable_scope('branch3x3'):
branch3x3 = ops.conv2d(net, 384, [1, 1])
branch3x3 = tf.concat(3, [
ops.conv2d(branch3x3, 384, [1, 3]),
ops.conv2d(branch3x3, 384, [3, 1])
])
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 448, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 384, [3, 3])
branch3x3dbl = tf.concat(3, [
ops.conv2d(branch3x3dbl, 384, [1, 3]),
ops.conv2d(branch3x3dbl, 384, [3, 1])
])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(
3, [branch1x1, branch3x3, branch3x3dbl, branch_pool])
end_points['mixed_8x8x2048a'] = net
# mixed_10: 8 x 8 x 2048.
with tf.variable_scope('mixed_8x8x2048b'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 320, [1, 1])
with tf.variable_scope('branch3x3'):
branch3x3 = ops.conv2d(net, 384, [1, 1])
branch3x3 = tf.concat(3, [
ops.conv2d(branch3x3, 384, [1, 3]),
ops.conv2d(branch3x3, 384, [3, 1])
])
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 448, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 384, [3, 3])
branch3x3dbl = tf.concat(3, [
ops.conv2d(branch3x3dbl, 384, [1, 3]),
ops.conv2d(branch3x3dbl, 384, [3, 1])
])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(
3, [branch1x1, branch3x3, branch3x3dbl, branch_pool])
end_points['mixed_8x8x2048b'] = net
# Final pooling and prediction
with tf.variable_scope('logits'):
shape = net.get_shape()
net = ops.avg_pool(net,
shape[1:3],
padding='VALID',
scope='pool')
# 1 x 1 x 2048
net = ops.dropout(net, dropout_keep_prob, scope='dropout')
net = ops.flatten(net, scope='flatten')
# 2048
logits = ops.fc(net,
num_classes,
activation=None,
scope='logits',
restore=restore_logits)
# 1000
end_points['logits'] = logits
end_points['predictions'] = tf.nn.softmax(
logits, name='predictions')
return [logits, aux_logits]
def inference(images, num_classes, scope):
with tf.op_scope([images], scope):
with scopes.arg_scope([ops.conv2d, ops.fc, ops.dropout], is_training=True):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(
images,
num_filters_out=64,
kernel_size=[11, 11],
stride=4,
stddev=1e-1)
with tf.variable_scope('pool1'):
pool1 = ops.max_pool(
conv1,
kernel_size=3,
stride=2,
padding='VALID')
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(
pool1,
num_filters_out=192,
kernel_size=1,
stride=1,
stddev=1e-1)
with tf.variable_scope('pool2'):
pool2 = ops.max_pool(
conv2,
kernel_size=3,
stride=2,
padding='VALID')
with tf.variable_scope('conv3'):
conv3 = ops.conv2d(
pool2,
num_filters_out=384,
kernel_size=3,
stride=1,
stddev=1e-1)
with tf.variable_scope('conv4'):
conv4 = ops.conv2d(
conv3,
num_filters_out=256,
kernel_size=3,
stride=1,
stddev=1e-1)
with tf.variable_scope('conv5'):
conv5 = ops.conv2d(
conv4,
num_filters_out=256,
kernel_size=3,
stride=1,
stddev=1e-1)
with tf.variable_scope('pool3'):
pool3 = ops.max_pool(
conv5,
kernel_size=3,
stride=2,
padding='VALID')
flattened = ops.flatten(pool3, scope='flatten')
with tf.variable_scope('fc1'):
fc1 = ops.fc(
flattened,
num_units_out=4096)
with tf.variable_scope('fc2'):
fc2 = ops.fc(
fc1,
num_units_out=4096)
with tf.variable_scope('fc3'):
fc3 = ops.fc(
fc2,
activation=None,
num_units_out=num_classes)
return fc3
def inference(inputs,
dropout_keep_prob=0.8,
num_classes=1000,
is_training=True,
restore_logits=True,
scope=''):
"""Latest Inception from http://arxiv.org/abs/1512.00567.
"Rethinking the Inception Architecture for Computer Vision"
Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens,
Zbigniew Wojna
Args:
inputs: a tensor of size [batch_size, height, width, channels].
dropout_keep_prob: dropout keep_prob.
num_classes: number of predicted classes.
is_training: whether is training or not.
restore_logits: whether or not the logits layers should be restored.
Useful for fine-tuning a model with different num_classes.
scope: Optional scope for op_scope.
Returns:
a list containing 'logits', 'aux_logits' Tensors.
"""
# end_points will collect relevant activations for external use, for example
# summaries or losses.
end_points = {}
batch_norm_params = {
# Decay for the batch_norm moving averages.
'decay': 0.9997,
# epsilon to prevent 0s in variance.
'epsilon': 0.001,
}
with tf.op_scope([inputs], scope, 'inception_v3'):
with scopes.arg_scope([ops.conv2d], batch_norm_params=batch_norm_params):
with scopes.arg_scope([ops.conv2d, ops.max_pool, ops.avg_pool],
stride=1, padding='VALID'):
# 299 x 299 x 3
end_points['conv0'] = ops.conv2d(inputs, 32, [3, 3], stride=2,
scope='conv0')
# 149 x 149 x 32
end_points['conv1'] = ops.conv2d(end_points['conv0'], 32, [3, 3],
scope='conv1')
# 147 x 147 x 32
end_points['conv2'] = ops.conv2d(end_points['conv1'], 64, [3, 3],
padding='SAME', scope='conv2')
# 147 x 147 x 64
end_points['pool1'] = ops.max_pool(end_points['conv2'], [3, 3],
stride=2, scope='pool1')
# 73 x 73 x 64
end_points['conv3'] = ops.conv2d(end_points['pool1'], 80, [1, 1],
scope='conv3')
# 73 x 73 x 80.
end_points['conv4'] = ops.conv2d(end_points['conv3'], 192, [3, 3],
scope='conv4')
# 71 x 71 x 192.
end_points['pool2'] = ops.max_pool(end_points['conv4'], [3, 3],
stride=2, scope='pool2')
# 35 x 35 x 192.
net = end_points['pool2']
# Inception blocks
with scopes.arg_scope([ops.conv2d, ops.max_pool, ops.avg_pool],
stride=1, padding='SAME'):
# mixed: 35 x 35 x 256.
with tf.variable_scope('mixed_35x35x256a'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 64, [1, 1])
with tf.variable_scope('branch5x5'):
branch5x5 = ops.conv2d(net, 48, [1, 1])
branch5x5 = ops.conv2d(branch5x5, 64, [5, 5])
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 64, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 32, [1, 1])
net = tf.concat(3, [branch1x1, branch5x5, branch3x3dbl, branch_pool])
end_points['mixed_35x35x256a'] = net
# mixed_1: 35 x 35 x 288.
with tf.variable_scope('mixed_35x35x288a'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 64, [1, 1])
with tf.variable_scope('branch5x5'):
branch5x5 = ops.conv2d(net, 48, [1, 1])
branch5x5 = ops.conv2d(branch5x5, 64, [5, 5])
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 64, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 64, [1, 1])
net = tf.concat(3, [branch1x1, branch5x5, branch3x3dbl, branch_pool])
end_points['mixed_35x35x288a'] = net
# mixed_2: 35 x 35 x 288.
with tf.variable_scope('mixed_35x35x288b'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 64, [1, 1])
with tf.variable_scope('branch5x5'):
branch5x5 = ops.conv2d(net, 48, [1, 1])
branch5x5 = ops.conv2d(branch5x5, 64, [5, 5])
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 64, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 64, [1, 1])
net = tf.concat(3, [branch1x1, branch5x5, branch3x3dbl, branch_pool])
end_points['mixed_35x35x288b'] = net
# mixed_3: 17 x 17 x 768.
with tf.variable_scope('mixed_17x17x768a'):
with tf.variable_scope('branch3x3'):
branch3x3 = ops.conv2d(net, 384, [3, 3], stride=2, padding='VALID')
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 64, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3])
branch3x3dbl = ops.conv2d(branch3x3dbl, 96, [3, 3],
stride=2, padding='VALID')
with tf.variable_scope('branch_pool'):
branch_pool = ops.max_pool(net, [3, 3], stride=2, padding='VALID')
net = tf.concat(3, [branch3x3, branch3x3dbl, branch_pool])
end_points['mixed_17x17x768a'] = net
# mixed4: 17 x 17 x 768.
with tf.variable_scope('mixed_17x17x768b'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 192, [1, 1])
with tf.variable_scope('branch7x7'):
branch7x7 = ops.conv2d(net, 128, [1, 1])
branch7x7 = ops.conv2d(branch7x7, 128, [1, 7])
branch7x7 = ops.conv2d(branch7x7, 192, [7, 1])
with tf.variable_scope('branch7x7dbl'):
branch7x7dbl = ops.conv2d(net, 128, [1, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 128, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 128, [1, 7])
branch7x7dbl = ops.conv2d(branch7x7dbl, 128, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [1, 7])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(3, [branch1x1, branch7x7, branch7x7dbl, branch_pool])
end_points['mixed_17x17x768b'] = net
# mixed_5: 17 x 17 x 768.
with tf.variable_scope('mixed_17x17x768c'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 192, [1, 1])
with tf.variable_scope('branch7x7'):
branch7x7 = ops.conv2d(net, 160, [1, 1])
branch7x7 = ops.conv2d(branch7x7, 160, [1, 7])
branch7x7 = ops.conv2d(branch7x7, 192, [7, 1])
with tf.variable_scope('branch7x7dbl'):
branch7x7dbl = ops.conv2d(net, 160, [1, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [1, 7])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [1, 7])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(3, [branch1x1, branch7x7, branch7x7dbl, branch_pool])
end_points['mixed_17x17x768c'] = net
# mixed_6: 17 x 17 x 768.
with tf.variable_scope('mixed_17x17x768d'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 192, [1, 1])
with tf.variable_scope('branch7x7'):
branch7x7 = ops.conv2d(net, 160, [1, 1])
branch7x7 = ops.conv2d(branch7x7, 160, [1, 7])
branch7x7 = ops.conv2d(branch7x7, 192, [7, 1])
with tf.variable_scope('branch7x7dbl'):
branch7x7dbl = ops.conv2d(net, 160, [1, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [1, 7])
branch7x7dbl = ops.conv2d(branch7x7dbl, 160, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [1, 7])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(3, [branch1x1, branch7x7, branch7x7dbl, branch_pool])
end_points['mixed_17x17x768d'] = net
# mixed_7: 17 x 17 x 768.
with tf.variable_scope('mixed_17x17x768e'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 192, [1, 1])
with tf.variable_scope('branch7x7'):
branch7x7 = ops.conv2d(net, 192, [1, 1])
branch7x7 = ops.conv2d(branch7x7, 192, [1, 7])
branch7x7 = ops.conv2d(branch7x7, 192, [7, 1])
with tf.variable_scope('branch7x7dbl'):
branch7x7dbl = ops.conv2d(net, 192, [1, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [1, 7])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [7, 1])
branch7x7dbl = ops.conv2d(branch7x7dbl, 192, [1, 7])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(3, [branch1x1, branch7x7, branch7x7dbl, branch_pool])
end_points['mixed_17x17x768e'] = net
# Auxiliary Head logits
aux_logits = tf.identity(end_points['mixed_17x17x768e'])
with tf.variable_scope('aux_logits'):
aux_logits = ops.avg_pool(aux_logits, [5, 5], stride=3,
padding='VALID')
aux_logits = ops.conv2d(aux_logits, 128, [1, 1], scope='proj')
# Shape of feature map before the final layer.
shape = aux_logits.get_shape()
aux_logits = ops.conv2d(aux_logits, 768, shape[1:3], stddev=0.01,
padding='VALID')
aux_logits = ops.flatten(aux_logits)
aux_logits = ops.fc(aux_logits, num_classes, activation=None,
stddev=0.001, restore=restore_logits)
end_points['aux_logits'] = aux_logits
# mixed_8: 8 x 8 x 1280.
# Note that the scope below is not changed to not void previous
# checkpoints.
# (TODO) Fix the scope when appropriate.
with tf.variable_scope('mixed_17x17x1280a'):
with tf.variable_scope('branch3x3'):
branch3x3 = ops.conv2d(net, 192, [1, 1])
branch3x3 = ops.conv2d(branch3x3, 320, [3, 3], stride=2,
padding='VALID')
with tf.variable_scope('branch7x7x3'):
branch7x7x3 = ops.conv2d(net, 192, [1, 1])
branch7x7x3 = ops.conv2d(branch7x7x3, 192, [1, 7])
branch7x7x3 = ops.conv2d(branch7x7x3, 192, [7, 1])
branch7x7x3 = ops.conv2d(branch7x7x3, 192, [3, 3],
stride=2, padding='VALID')
with tf.variable_scope('branch_pool'):
branch_pool = ops.max_pool(net, [3, 3], stride=2, padding='VALID')
net = tf.concat(3, [branch3x3, branch7x7x3, branch_pool])
end_points['mixed_17x17x1280a'] = net
# mixed_9: 8 x 8 x 2048.
with tf.variable_scope('mixed_8x8x2048a'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 320, [1, 1])
with tf.variable_scope('branch3x3'):
branch3x3 = ops.conv2d(net, 384, [1, 1])
branch3x3 = tf.concat(3, [ops.conv2d(branch3x3, 384, [1, 3]),
ops.conv2d(branch3x3, 384, [3, 1])])
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 448, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 384, [3, 3])
branch3x3dbl = tf.concat(3, [ops.conv2d(branch3x3dbl, 384, [1, 3]),
ops.conv2d(branch3x3dbl, 384, [3, 1])])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(3, [branch1x1, branch3x3, branch3x3dbl, branch_pool])
end_points['mixed_8x8x2048a'] = net
# mixed_10: 8 x 8 x 2048.
with tf.variable_scope('mixed_8x8x2048b'):
with tf.variable_scope('branch1x1'):
branch1x1 = ops.conv2d(net, 320, [1, 1])
with tf.variable_scope('branch3x3'):
branch3x3 = ops.conv2d(net, 384, [1, 1])
branch3x3 = tf.concat(3, [ops.conv2d(branch3x3, 384, [1, 3]),
ops.conv2d(branch3x3, 384, [3, 1])])
with tf.variable_scope('branch3x3dbl'):
branch3x3dbl = ops.conv2d(net, 448, [1, 1])
branch3x3dbl = ops.conv2d(branch3x3dbl, 384, [3, 3])
branch3x3dbl = tf.concat(3, [ops.conv2d(branch3x3dbl, 384, [1, 3]),
ops.conv2d(branch3x3dbl, 384, [3, 1])])
with tf.variable_scope('branch_pool'):
branch_pool = ops.avg_pool(net, [3, 3])
branch_pool = ops.conv2d(branch_pool, 192, [1, 1])
net = tf.concat(3, [branch1x1, branch3x3, branch3x3dbl, branch_pool])
end_points['mixed_8x8x2048b'] = net
# Final pooling and prediction
with tf.variable_scope('logits'):
shape = net.get_shape()
net = ops.avg_pool(net, shape[1:3], padding='VALID', scope='pool')
# 1 x 1 x 2048
net = ops.dropout(net, dropout_keep_prob, scope='dropout')
net = ops.flatten(net, scope='flatten')
# 2048
logits = ops.fc(net, num_classes, activation=None, scope='logits',
restore=restore_logits)
# 1000
end_points['logits'] = logits
end_points['predictions'] = tf.nn.softmax(logits, name='predictions')
return [logits, aux_logits]
def inference(inputs, num_classes, scope):
with tf.op_scope([inputs], scope, 'vgg19'):
with scopes.arg_scope(
[ops.conv2d, ops.fc, ops.batch_norm, ops.dropout],
is_training=True):
# block 1
with tf.variable_scope('block1'):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(inputs,
num_filters_out=64,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(conv1,
num_filters_out=64,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('pool'):
pool = ops.max_pool(conv2,
kernel_size=2,
stride=2,
padding='VALID')
# block 2
with tf.variable_scope('block2'):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(pool,
num_filters_out=128,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(conv1,
num_filters_out=128,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('pool'):
pool = ops.max_pool(conv2,
kernel_size=2,
stride=2,
padding='VALID')
# block 3
with tf.variable_scope('block3'):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(pool,
num_filters_out=256,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(conv1,
num_filters_out=256,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv3'):
conv3 = ops.conv2d(conv2,
num_filters_out=256,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('pool'):
pool = ops.max_pool(conv3,
kernel_size=2,
stride=2,
padding='VALID')
# block 4
with tf.variable_scope('block4'):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(pool,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(conv1,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv3'):
conv3 = ops.conv2d(conv2,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('pool'):
pool = ops.max_pool(conv3,
kernel_size=2,
stride=2,
padding='VALID')
# block 5
with tf.variable_scope('block5'):
with tf.variable_scope('conv1'):
conv1 = ops.conv2d(pool,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv2'):
conv2 = ops.conv2d(conv1,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('conv3'):
conv3 = ops.conv2d(conv2,
num_filters_out=512,
kernel_size=[3, 3],
stride=1,
stddev=1e-1)
with tf.variable_scope('pool'):
pool = ops.max_pool(conv3,
kernel_size=2,
stride=2,
padding='VALID')
flattened = ops.flatten(pool, scope='flatten')
# fc
with tf.variable_scope('fc1'):
fc1 = ops.fc(flattened, num_units_out=4096)
with tf.variable_scope('fc2'):
fc2 = ops.fc(fc1, num_units_out=4096)
with tf.variable_scope('fc3'):
fc3 = ops.fc(fc2, activation=None, num_units_out=num_classes)
return fc3
