def depthmotion_block_demon_original(image_pair,
image2_2,
prev_flow2,
prev_flowconf2,
prev_rotation=None,
prev_translation=None,
intrinsics=None,
data_format='channels_first'):
"""Creates a depth and motion network
image_pair: Tensor
Image pair concatenated along the channel axis.
The tensor format is NCHW with C == 6.
image2_2: Tensor
Second image at resolution level 2
prev_flow2: Tensor
The output of the flow network. Contains only the flow (2 channels)
prev_flowconf2: Tensor
The output of the flow network. Contains flow and flow confidence (4 channels)
prev_rotation: Tensor
The previously predicted rotation.
prev_translaion: Tensor
The previously predicted translation.
intrinsics: Tensor
Tensor with the intrinsic camera parameters
Only required if prev_rotation and prev_translation is not None.
Returns a dictionary with the predictions for depth, normals and motion
"""
conv_params = {'data_format': data_format}
fc_params = {}
# contracting part
conv1 = convrelu2_caffe_padding(name='conv1',
inputs=image_pair,
num_outputs=32,
kernel_size=9,
stride=2,
**conv_params)
conv2 = convrelu2_caffe_padding(name='conv2',
inputs=conv1,
num_outputs=32,
kernel_size=7,
stride=2,
**conv_params)
# create extra inputs
if data_format == 'channels_first':
image2_2_warped = sops.warp2d(image2_2,
prev_flow2,
normalized=True,
border_mode='value')
else:
prev_flow2_nchw = convert_NHWC_to_NCHW(prev_flow2)
image2_2_warped = convert_NCHW_to_NHWC(
sops.warp2d(convert_NHWC_to_NCHW(image2_2),
prev_flow2_nchw,
normalized=True,
border_mode='value'))
extra_inputs = [image2_2_warped, prev_flowconf2]
if (not prev_rotation is None) and (not prev_translation is None) and (
not intrinsics is None):
if data_format == 'channels_first':
depth_from_flow = sops.flow_to_depth(
flow=prev_flow2,
intrinsics=intrinsics,
rotation=prev_rotation,
translation=prev_translation,
normalized_flow=True,
inverse_depth=True,
)
else:
depth_from_flow = convert_NCHW_to_NHWC(
sops.flow_to_depth(
flow=prev_flow2_nchw,
intrinsics=intrinsics,
rotation=prev_rotation,
translation=prev_translation,
normalized_flow=True,
inverse_depth=True,
))
extra_inputs.append(depth_from_flow)
concat_extra_inputs = tf.stop_gradient(
tf.concat(extra_inputs,
axis=1 if data_format == 'channels_first' else 3))
conv_extra_inputs = convrelu2_caffe_padding(name='conv2_extra_inputs',
inputs=concat_extra_inputs,
num_outputs=32,
kernel_size=3,
stride=1,
**conv_params)
conv2_concat = tf.concat((conv2, conv_extra_inputs),
axis=1 if data_format == 'channels_first' else 3)
conv2_1 = convrelu2_caffe_padding(name='conv2_1',
inputs=conv2_concat,
num_outputs=64,
kernel_size=3,
stride=1,
**conv_params)
tf.add_to_collection('endpoints', conv2_1)
conv3 = convrelu2_caffe_padding(name='conv3',
inputs=conv2_1,
num_outputs=128,
kernel_size=5,
stride=2,
**conv_params)
conv3_1 = convrelu2_caffe_padding(name='conv3_1',
inputs=conv3,
num_outputs=128,
kernel_size=3,
stride=1,
**conv_params)
tf.add_to_collection('endpoints', conv3_1)
conv4 = convrelu2_caffe_padding(name='conv4',
inputs=conv3_1,
num_outputs=256,
kernel_size=5,
stride=2,
**conv_params)
conv4_1 = convrelu2_caffe_padding(name='conv4_1',
inputs=conv4,
num_outputs=256,
kernel_size=3,
stride=1,
**conv_params)
tf.add_to_collection('endpoints', conv4_1)
conv5 = convrelu2_caffe_padding(name='conv5',
inputs=conv4_1,
num_outputs=512,
kernel_size=3,
stride=2,
**conv_params)
conv5_1 = convrelu2_caffe_padding(name='conv5_1',
inputs=conv5,
num_outputs=512,
kernel_size=3,
stride=1,
**conv_params)
tf.add_to_collection('endpoints', conv5_1)
# motion prediction part
motion_conv1 = convrelu_caffe_padding(
name='motion_conv1',
inputs=conv5_1,
num_outputs=128,
kernel_size=3,
strides=1,
**conv_params,
)
tf.add_to_collection('endpoints', motion_conv1)
if data_format == 'channels_last':
motion_conv1 = convert_NHWC_to_NCHW(motion_conv1)
motion_fc1 = tf.layers.dense(
name='motion_fc1',
inputs=tf.contrib.layers.flatten(motion_conv1),
units=1024,
activation=myLeakyRelu,
**fc_params,
)
motion_fc2 = tf.layers.dense(
name='motion_fc2',
inputs=motion_fc1,
units=128,
activation=myLeakyRelu,
**fc_params,
)
predict_motion_scale = tf.layers.dense(
name='motion_fc3',
inputs=motion_fc2,
units=7,
activation=None,
**fc_params,
)
predict_rotation, predict_translation, predict_scale = tf.split(
value=predict_motion_scale, num_or_size_splits=[3, 3, 1], axis=1)
# expanding part
with tf.variable_scope('refine4'):
concat4 = _refine_caffe_padding(
inp=conv5_1,
num_outputs=256,
features_direct=conv4_1,
**conv_params,
)
with tf.variable_scope('refine3'):
concat3 = _refine_caffe_padding(
inp=concat4,
num_outputs=128,
features_direct=conv3_1,
**conv_params,
)
with tf.variable_scope('refine2'):
concat2 = _refine_caffe_padding(
inp=concat3,
num_outputs=64,
features_direct=conv2_1,
**conv_params,
)
with tf.variable_scope('predict_depthnormal2'):
predict_depth2, predict_normal2 = _predict_depthnormal_caffe_padding(
concat2, predicted_scale=predict_scale, **conv_params)
return {
'predict_depth2': predict_depth2,
'predict_normal2': predict_normal2,
'predict_rotation': predict_rotation,
'predict_translation': predict_translation,
'predict_scale': predict_scale,
}
def depthmotion_block(image_pair,
image2_2,
prev_flow2,
prev_flowconf2,
prev_rotation=None,
prev_translation=None,
intrinsics=None,
data_format='channels_first',
kernel_regularizer=None):
"""Creates a depth and motion network
image_pair: Tensor
Image pair concatenated along the channel axis.
The tensor format is NCHW with C == 6.
image2_2: Tensor
Second image at resolution level 2
prev_flow2: Tensor
The output of the flow network. Contains only the flow (2 channels)
prev_flowconf2: Tensor
The output of the flow network. Contains flow and flow confidence (4 channels)
prev_rotation: Tensor
The previously predicted rotation.
prev_translaion: Tensor
The previously predicted translation.
intrinsics: Tensor
Tensor with the intrinsic camera parameters
Only required if prev_rotation and prev_translation is not None.
Returns a dictionary with the predictions for depth, normals and motion
"""
conv_params = {
'data_format': data_format,
'kernel_regularizer': kernel_regularizer
}
fc_params = {}
# contracting part
conv1 = convrelu2(name='conv1',
inputs=image_pair,
num_outputs=(24, 32),
kernel_size=9,
stride=2,
**conv_params)
conv2 = convrelu2(name='conv2',
inputs=conv1,
num_outputs=32,
kernel_size=7,
stride=2,
**conv_params)
# create extra inputs
if data_format == 'channels_first':
image2_2_warped = sops.warp2d(image2_2,
prev_flow2,
normalized=True,
border_mode='value')
else:
prev_flow2_nchw = convert_NHWC_to_NCHW(prev_flow2)
image2_2_warped = convert_NCHW_to_NHWC(
sops.warp2d(convert_NHWC_to_NCHW(image2_2),
prev_flow2_nchw,
normalized=True,
border_mode='value'))
extra_inputs = [image2_2_warped, prev_flowconf2]
if (not prev_rotation is None) and (not prev_translation is None) and (
not intrinsics is None):
if data_format == 'channels_first':
depth_from_flow = sops.flow_to_depth2(
flow=prev_flow2,
intrinsics=intrinsics,
rotation=prev_rotation,
translation=prev_translation,
normalized_flow=True,
inverse_depth=True,
)
else:
depth_from_flow = convert_NCHW_to_NHWC(
sops.flow_to_depth2(
flow=prev_flow2_nchw,
intrinsics=intrinsics,
rotation=prev_rotation,
translation=prev_translation,
normalized_flow=True,
inverse_depth=True,
))
depth_from_flow = tf.clip_by_value(depth_from_flow, 0.0, 50.0)
extra_inputs.append(depth_from_flow)
concat_extra_inputs = tf.stop_gradient(
tf.concat(extra_inputs,
axis=1 if data_format == 'channels_first' else 3))
conv_extra_inputs = convrelu2(name='conv2_extra_inputs',
inputs=concat_extra_inputs,
num_outputs=32,
kernel_size=3,
stride=1,
**conv_params)
conv2_concat = tf.concat((conv2, conv_extra_inputs),
axis=1 if data_format == 'channels_first' else 3)
conv2_1 = convrelu2(name='conv2_1',
inputs=conv2_concat,
num_outputs=64,
kernel_size=3,
stride=1,
**conv_params)
conv3 = convrelu2(name='conv3',
inputs=conv2_1,
num_outputs=(96, 128),
kernel_size=5,
stride=2,
**conv_params)
conv3_1 = convrelu2(name='conv3_1',
inputs=conv3,
num_outputs=128,
kernel_size=3,
stride=1,
**conv_params)
conv4 = convrelu2(name='conv4',
inputs=conv3_1,
num_outputs=(192, 256),
kernel_size=5,
stride=2,
**conv_params)
conv4_1 = convrelu2(name='conv4_1',
inputs=conv4,
num_outputs=256,
kernel_size=3,
stride=1,
**conv_params)
conv5 = convrelu2(name='conv5',
inputs=conv4_1,
num_outputs=384,
kernel_size=3,
stride=2,
**conv_params)
conv5_1 = convrelu2(name='conv5_1',
inputs=conv5,
num_outputs=384,
kernel_size=3,
stride=1,
**conv_params)
dense_slice_shape = conv5_1.get_shape().as_list()
if data_format == 'channels_first':
dense_slice_shape[1] = 96
else:
dense_slice_shape[-1] = 96
units = 1
for i in range(1, len(dense_slice_shape)):
units *= dense_slice_shape[i]
dense5 = tf.layers.dense(tf.contrib.layers.flatten(
tf.slice(conv5_1, [0, 0, 0, 0], dense_slice_shape)),
units=units,
activation=myLeakyRelu,
kernel_initializer=default_weights_initializer(),
kernel_regularizer=kernel_regularizer,
name='dense5')
print(dense5)
conv5_1_dense5 = tf.concat(
(conv5_1, tf.reshape(dense5, dense_slice_shape)),
axis=1 if data_format == 'channels_first' else 3)
# motion prediction part
motion_conv3 = convrelu2(name='motion_conv3',
inputs=conv2_1,
num_outputs=64,
kernel_size=5,
stride=2,
**conv_params)
motion_conv4 = convrelu2(name='motion_conv4',
inputs=motion_conv3,
num_outputs=64,
kernel_size=5,
stride=2,
**conv_params)
motion_conv5a = convrelu2(name='motion_conv5a',
inputs=motion_conv4,
num_outputs=64,
kernel_size=3,
stride=2,
**conv_params)
motion_conv5b = convrelu(
name='motion_conv5b',
inputs=conv5_1_dense5,
num_outputs=64,
kernel_size=3,
strides=1,
**conv_params,
)
motion_conv5_1 = tf.concat(
(motion_conv5a, motion_conv5b),
axis=1 if data_format == 'channels_first' else 3)
if data_format == 'channels_last':
motion_conv5_1 = convert_NHWC_to_NCHW(motion_conv5_1)
motion_fc1 = tf.layers.dense(
name='motion_fc1',
inputs=tf.contrib.layers.flatten(motion_conv5_1),
units=1024,
activation=myLeakyRelu,
kernel_regularizer=kernel_regularizer,
**fc_params,
)
motion_fc2 = tf.layers.dense(
name='motion_fc2',
inputs=motion_fc1,
units=128,
activation=myLeakyRelu,
kernel_regularizer=kernel_regularizer,
**fc_params,
)
predict_motion_scale = tf.layers.dense(
name='motion_fc3',
inputs=motion_fc2,
units=7,
activation=None,
kernel_regularizer=kernel_regularizer,
**fc_params,
)
predict_rotation, predict_translation, predict_scale = tf.split(
value=predict_motion_scale, num_or_size_splits=[3, 3, 1], axis=1)
# expanding part
with tf.variable_scope('refine4'):
concat4 = _refine(
inp=conv5_1,
num_outputs=256,
features_direct=conv4_1,
**conv_params,
)
with tf.variable_scope('refine3'):
concat3 = _refine(
inp=concat4,
num_outputs=128,
features_direct=conv3_1,
**conv_params,
)
with tf.variable_scope('refine2'):
concat2 = _refine(
inp=concat3,
num_outputs=64,
features_direct=conv2_1,
**conv_params,
)
with tf.variable_scope('predict_depthnormal2'):
predict_depth2, predict_normal2 = _predict_depthnormal(
concat2, predicted_scale=predict_scale, **conv_params)
return {
'predict_depth2': predict_depth2,
'predict_normal2': predict_normal2,
'predict_rotation': predict_rotation,
'predict_translation': predict_translation,
'predict_scale': predict_scale,
}
def flow_block_demon_original(image_pair,
image2_2=None,
intrinsics=None,
prev_predictions=None,
data_format='channels_first'):
"""Creates a flow network
image_pair: Tensor
Image pair concatenated along the channel axis.
image2_2: Tensor
Second image at resolution level 2 (downsampled two times)
intrinsics: Tensor
The normalized intrinsic parameters
prev_predictions: dict of Tensor
Predictions from the previous depth block
Returns a dict with the predictions
"""
conv_params = {'data_format': data_format}
# contracting part
conv1 = convrelu2_caffe_padding(name='conv1',
inputs=image_pair,
num_outputs=32,
kernel_size=9,
stride=2,
**conv_params)
if prev_predictions is None:
conv2 = convrelu2_caffe_padding(name='conv2',
inputs=conv1,
num_outputs=64,
kernel_size=7,
stride=2,
**conv_params)
conv2_1 = convrelu2_caffe_padding(name='conv2_1',
inputs=conv2,
num_outputs=64,
kernel_size=3,
stride=1,
**conv_params)
tf.add_to_collection('endpoints', conv2_1)
else:
conv2 = convrelu2_caffe_padding(name='conv2',
inputs=conv1,
num_outputs=32,
kernel_size=7,
stride=2,
**conv_params)
# create warped input
if data_format == 'channels_first':
prev_depth_nchw = prev_predictions['predict_depth2']
else:
prev_depth_nchw = convert_NHWC_to_NCHW(
prev_predictions['predict_depth2'])
_flow_from_depth_motion = sops.depth_to_flow(
intrinsics=intrinsics,
depth=prev_depth_nchw,
rotation=prev_predictions['predict_rotation'],
translation=prev_predictions['predict_translation'],
inverse_depth=True,
normalize_flow=True,
)
# set flow vectors to zero if the motion is too large.
# this also eliminates nan values which can be produced by very bad camera parameters
flow_from_depth_motion_norm = tf.norm(_flow_from_depth_motion,
axis=1,
keep_dims=True)
flow_from_depth_motion_norm = tf.concat(
(flow_from_depth_motion_norm, flow_from_depth_motion_norm), axis=1)
tmp_zeros = tf.zeros_like(_flow_from_depth_motion, dtype=tf.float32)
flow_from_depth_motion = tf.where(flow_from_depth_motion_norm < 1.0,
_flow_from_depth_motion, tmp_zeros)
image2_2_warped = sops.warp2d(
input=image2_2 if data_format == 'channels_first' else
convert_NHWC_to_NCHW(image2_2),
displacements=flow_from_depth_motion,
normalized=True,
border_mode='value',
)
if data_format == 'channels_last':
flow_from_depth_motion = convert_NCHW_to_NHWC(
flow_from_depth_motion)
image2_2_warped = convert_NCHW_to_NHWC(image2_2_warped)
extra_inputs = (image2_2_warped, flow_from_depth_motion,
prev_predictions['predict_depth2'],
prev_predictions['predict_normal2'])
# stop gradient here
extra_inputs_concat = tf.stop_gradient(
tf.concat(extra_inputs,
axis=1 if data_format == 'channels_first' else 3))
conv_extra_inputs = convrelu2_caffe_padding(name='conv2_extra_inputs',
inputs=extra_inputs_concat,
num_outputs=32,
kernel_size=3,
stride=1,
**conv_params)
conv2_concat = tf.concat(
(conv2, conv_extra_inputs),
axis=1 if data_format == 'channels_first' else 3)
conv2_1 = convrelu2_caffe_padding(name='conv2_1',
inputs=conv2_concat,
num_outputs=64,
kernel_size=3,
stride=1,
**conv_params)
conv3 = convrelu2_caffe_padding(name='conv3',
inputs=conv2_1,
num_outputs=128,
kernel_size=5,
stride=2,
**conv_params)
conv3_1 = convrelu2_caffe_padding(name='conv3_1',
inputs=conv3,
num_outputs=128,
kernel_size=3,
stride=1,
**conv_params)
tf.add_to_collection('endpoints', conv3_1)
conv4 = convrelu2_caffe_padding(name='conv4',
inputs=conv3_1,
num_outputs=256,
kernel_size=5,
stride=2,
**conv_params)
conv4_1 = convrelu2_caffe_padding(name='conv4_1',
inputs=conv4,
num_outputs=256,
kernel_size=3,
stride=1,
**conv_params)
tf.add_to_collection('endpoints', conv4_1)
conv5 = convrelu2_caffe_padding(name='conv5',
inputs=conv4_1,
num_outputs=512,
kernel_size=5,
stride=2,
**conv_params)
conv5_1 = convrelu2_caffe_padding(name='conv5_1',
inputs=conv5,
num_outputs=512,
kernel_size=3,
stride=1,
**conv_params)
tf.add_to_collection('endpoints', conv5_1)
# expanding part
with tf.variable_scope('predict_flow5'):
predict_flowconf5 = _predict_flow_caffe_padding(
conv5_1, predict_confidence=True, **conv_params)
tf.add_to_collection('endpoints', predict_flowconf5)
with tf.variable_scope('upsample_flow5to4'):
predict_flowconf5to4 = _upsample_prediction(predict_flowconf5, 2,
**conv_params)
with tf.variable_scope('refine4'):
concat4 = _refine_caffe_padding(
inp=conv5_1,
num_outputs=256,
upsampled_prediction=predict_flowconf5to4,
features_direct=conv4_1,
**conv_params,
)
with tf.variable_scope('refine3'):
concat3 = _refine_caffe_padding(
inp=concat4,
num_outputs=128,
features_direct=conv3_1,
**conv_params,
)
with tf.variable_scope('refine2'):
concat2 = _refine_caffe_padding(
inp=concat3,
num_outputs=64,
features_direct=conv2_1,
**conv_params,
)
with tf.variable_scope('predict_flow2'):
predict_flowconf2 = _predict_flow_caffe_padding(
concat2, predict_confidence=True, **conv_params)
tf.add_to_collection('endpoints', predict_flowconf2)
return {
'predict_flowconf5': predict_flowconf5,
'predict_flowconf2': predict_flowconf2
}
def flow_block(image_pair,
image2_2=None,
intrinsics=None,
prev_predictions=None,
data_format='channels_first',
kernel_regularizer=None):
"""Creates a flow network
image_pair: Tensor
Image pair concatenated along the channel axis.
image2_2: Tensor
Second image at resolution level 2 (downsampled two times)
intrinsics: Tensor
The normalized intrinsic parameters
prev_predictions: dict of Tensor
Predictions from the previous depth block
Returns a dict with the predictions
"""
conv_params = {
'data_format': data_format,
'kernel_regularizer': kernel_regularizer
}
# contracting part
conv1 = convrelu2(name='conv1',
inputs=image_pair,
num_outputs=(24, 32),
kernel_size=9,
stride=2,
**conv_params)
if prev_predictions is None:
conv2 = convrelu2(name='conv2',
inputs=conv1,
num_outputs=(48, 64),
kernel_size=7,
stride=2,
**conv_params)
conv2_1 = convrelu2(name='conv2_1',
inputs=conv2,
num_outputs=64,
kernel_size=3,
stride=1,
**conv_params)
else:
conv2 = convrelu2(name='conv2',
inputs=conv1,
num_outputs=32,
kernel_size=7,
stride=2,
**conv_params)
# create warped input
if data_format == 'channels_first':
prev_depth_nchw = prev_predictions['predict_depth2']
else:
prev_depth_nchw = convert_NHWC_to_NCHW(
prev_predictions['predict_depth2'])
_flow_from_depth_motion = sops.depth_to_flow(
intrinsics=intrinsics,
depth=prev_depth_nchw,
rotation=prev_predictions['predict_rotation'],
translation=prev_predictions['predict_translation'],
inverse_depth=True,
normalize_flow=True,
)
# set flow vectors to zero if the motion is too large.
# this also eliminates nan values which can be produced by very bad camera parameters
flow_from_depth_motion_norm = tf.norm(_flow_from_depth_motion,
axis=1,
keep_dims=True)
flow_from_depth_motion_norm = tf.concat(
(flow_from_depth_motion_norm, flow_from_depth_motion_norm), axis=1)
tmp_zeros = tf.zeros_like(_flow_from_depth_motion, dtype=tf.float32)
flow_from_depth_motion = tf.where(flow_from_depth_motion_norm < 1.0,
_flow_from_depth_motion, tmp_zeros)
image2_2_warped = sops.warp2d(
input=image2_2 if data_format == 'channels_first' else
convert_NHWC_to_NCHW(image2_2),
displacements=flow_from_depth_motion,
normalized=True,
border_mode='value',
)
if data_format == 'channels_last':
flow_from_depth_motion = convert_NCHW_to_NHWC(
flow_from_depth_motion)
image2_2_warped = convert_NCHW_to_NHWC(image2_2_warped)
extra_inputs = (image2_2_warped, flow_from_depth_motion,
prev_predictions['predict_depth2'],
prev_predictions['predict_normal2'])
# stop gradient here
extra_inputs_concat = tf.stop_gradient(
tf.concat(extra_inputs,
axis=1 if data_format == 'channels_first' else 3))
conv_extra_inputs = convrelu2(name='conv2_extra_inputs',
inputs=extra_inputs_concat,
num_outputs=32,
kernel_size=3,
stride=1,
**conv_params)
conv2_concat = tf.concat(
(conv2, conv_extra_inputs),
axis=1 if data_format == 'channels_first' else 3)
conv2_1 = convrelu2(name='conv2_1',
inputs=conv2_concat,
num_outputs=64,
kernel_size=3,
stride=1,
**conv_params)
conv3 = convrelu2(name='conv3',
inputs=conv2_1,
num_outputs=(96, 128),
kernel_size=5,
stride=2,
**conv_params)
conv3_1 = convrelu2(name='conv3_1',
inputs=conv3,
num_outputs=128,
kernel_size=3,
stride=1,
**conv_params)
conv4 = convrelu2(name='conv4',
inputs=conv3_1,
num_outputs=(192, 256),
kernel_size=5,
stride=2,
**conv_params)
conv4_1 = convrelu2(name='conv4_1',
inputs=conv4,
num_outputs=256,
kernel_size=3,
stride=1,
**conv_params)
conv5 = convrelu2(name='conv5',
inputs=conv4_1,
num_outputs=384,
kernel_size=5,
stride=2,
**conv_params)
conv5_1 = convrelu2(name='conv5_1',
inputs=conv5,
num_outputs=384,
kernel_size=3,
stride=1,
**conv_params)
dense_slice_shape = conv5_1.get_shape().as_list()
if data_format == 'channels_first':
dense_slice_shape[1] = 96
else:
dense_slice_shape[-1] = 96
units = 1
for i in range(1, len(dense_slice_shape)):
units *= dense_slice_shape[i]
dense5 = tf.layers.dense(tf.contrib.layers.flatten(
tf.slice(conv5_1, [0, 0, 0, 0], dense_slice_shape)),
units=units,
activation=myLeakyRelu,
kernel_initializer=default_weights_initializer(),
kernel_regularizer=kernel_regularizer,
name='dense5')
print(dense5)
conv5_1_dense5 = tf.concat(
(conv5_1, tf.reshape(dense5, dense_slice_shape)),
axis=1 if data_format == 'channels_first' else 3)
# expanding part
with tf.variable_scope('predict_flow5'):
predict_flowconf5 = _predict_flow(conv5_1_dense5,
predict_confidence=True,
**conv_params)
with tf.variable_scope('upsample_flow5to4'):
predict_flowconf5to4 = _upsample_prediction(predict_flowconf5, 2,
**conv_params)
with tf.variable_scope('refine4'):
concat4 = _refine(
inp=conv5_1_dense5,
num_outputs=256,
upsampled_prediction=predict_flowconf5to4,
features_direct=conv4_1,
**conv_params,
)
with tf.variable_scope('refine3'):
concat3 = _refine(
inp=concat4,
num_outputs=128,
features_direct=conv3_1,
**conv_params,
)
with tf.variable_scope('refine2'):
concat2 = _refine(
inp=concat3,
num_outputs=64,
features_direct=conv2_1,
**conv_params,
)
with tf.variable_scope('predict_flow2'):
predict_flowconf2 = _predict_flow(concat2,
predict_confidence=True,
**conv_params)
return {
'predict_flowconf5': predict_flowconf5,
'predict_flowconf2': predict_flowconf2
}