def inference(hypes, images, train=True):
"""Build the MNIST model up to where it may be used for inference.
Args:
images: Images placeholder, from inputs().
train: whether the network is used for train of inference
Returns:
softmax_linear: Output tensor with the computed logits.
"""
vgg16_npy_path = os.path.join(hypes['dirs']['data_dir'], "vgg16.npy")
vgg_fcn = fcn8_vgg.FCN8VGG(vgg16_npy_path=vgg16_npy_path)
num_classes = 2 # does not influence training what so ever
vgg_fcn.wd = hypes['wd']
vgg_fcn.build(images, train=train, num_classes=num_classes,
random_init_fc8=True)
vgg_dict = {'unpooled': vgg_fcn.conv5_3,
'deep_feat': vgg_fcn.pool5,
'deep_feat_channels': 512,
'early_feat': vgg_fcn.conv4_3,
'scored_feat': vgg_fcn.score_fr}
return vgg_dict
def inference(hypes, images, train=True):
"""Build the MNIST model up to where it may be used for inference.
Args:
images: Images placeholder, from inputs().
train: whether the network is used for train of inference
Returns:
softmax_linear: Output tensor with the computed logits.
"""
vgg16_npy_path = os.path.join(hypes['dirs']['data_dir'], "vgg16.npy")
vgg_fcn = fcn8_vgg.FCN8VGG(vgg16_npy_path=vgg16_npy_path)
num_classes = 3 # does not influence training what so ever
vgg_fcn.wd = hypes['wd']
vgg_fcn.build(images,
train=train,
num_classes=num_classes,
random_init_fc8=True)
if hypes['arch']['deep_feat'] == "pool5":
deep_feat = vgg_fcn.pool5
elif hypes['arch']['deep_feat'] == "fc7":
deep_feat = vgg_fcn.fc7
else:
raise NotImplementedError
vgg_dict = {'deep_feat': deep_feat, 'early_feat': vgg_fcn.conv4_3}
return vgg_dict
def inference(hypes, images, train=True):
"""Build the MNIST model up to where it may be used for inference.
Args:
images: Images placeholder, from inputs().
train: whether the network is used for train of inference
Returns:
softmax_linear: Output tensor with the computed logits.
"""
vgg16_npy_path = os.path.join(hypes['dirs']['data_dir'], "model_2D.pkl")
vgg_fcn = fcn8_vgg.FCN8VGG(vgg16_npy_path=vgg16_npy_path)
'''
num_classes does not influence training when KittiBox is used alone.
However, if you wish to use MultiNet with custom submodules,
e.g., if KittiSeg is customized for != 2 classes, this value must
reflect num_classes in KittiSeg, since MultiNet will try to share
variable score_fr/weights
'''
num_classes = 2
vgg_fcn.wd = hypes['wd']
vgg_fcn.build(images,
train=train,
num_classes=num_classes,
random_init_fc8=True)
if hypes['arch']['deep_feat'] == "pool5":
deep_feat = vgg_fcn.pool5
elif hypes['arch']['deep_feat'] == "fc7":
deep_feat = vgg_fcn.fc7
else:
raise NotImplementedError
vgg_dict = {
'deep_feat': deep_feat,
'early_feat': vgg_fcn.conv4_3,
'depth_early_feat': vgg_fcn.conv4_depth,
'depth_deep_feat': vgg_fcn.pool5_depth,
'location_early_feat': vgg_fcn.conv4_location,
'location_deep_feat': vgg_fcn.pool5_location,
'corner_early_feat': vgg_fcn.conv4_corner,
'corner_deep_feat': vgg_fcn.pool5_corner
}
return vgg_dict
def inference(hypes, images, train=True):
"""Build the MNIST model up to where it may be used for inference.
Args:
images: Images placeholder, from inputs().
train: whether the network is used for train of inference
Returns:
softmax_linear: Output tensor with the computed logits.
"""
vgg16_npy_path = os.path.join(hypes['dirs']['data_dir'], 'weights',
"vgg16.npy")
vgg_fcn = fcn8_vgg.FCN8VGG(vgg16_npy_path=vgg16_npy_path)
vgg_fcn.wd = hypes['wd']
num_classes = hypes['arch']['num_classes']
vgg_fcn.build(images,
train=train,
num_classes=num_classes,
random_init_fc8=True)
logits = {}
logits['images'] = images
if hypes['arch']['fcn_in'] == 'pool5':
logits['fcn_in'] = vgg_fcn.pool5
elif hypes['arch']['fcn_in'] == 'fc7':
logits['fcn_in'] = vgg_fcn.fc7
else:
raise NotImplementedError
logits['feed2'] = vgg_fcn.pool4
logits['feed4'] = vgg_fcn.pool3
logits['feed4'] = vgg_fcn.pool3
logits['fcn_logits'] = vgg_fcn.upscore32
logits['deep_feat'] = vgg_fcn.pool5
logits['early_feat'] = vgg_fcn.conv4_3
return logits
def inference(hypes, images, train=True):
"""Build the MNIST model up to where it may be used for inference.
Args:
images: Images placeholder, from inputs().
train: whether the network is used for train of inference
Returns:
softmax_linear: Output tensor with the computed logits.
"""
vgg16_npy_path = os.path.join(hypes['dirs']['data_dir'], "vgg16.npy")
vgg_fcn = fcn8_vgg.FCN8VGG(vgg16_npy_path=vgg16_npy_path)
vgg_fcn.wd = hypes['wd']
vgg_fcn.build(images, train=train, num_classes=2, random_init_fc8=True)
return vgg_fcn.upscore32
def inference(hypes, images, train=True):
"""Build the MNIST model up to where it may be used for inference.
Args:
images: Images placeholder, from inputs().
train: whether the network is used for train of inference
Returns:
softmax_linear: Output tensor with the computed logits.
"""
vgg16_npy_path = os.path.join(hypes['dirs']['data_dir'], 'weights',
"vgg16.npy")
vgg_fcn = fcn8_vgg.FCN8VGG(vgg16_npy_path=vgg16_npy_path)
vgg_fcn.wd = hypes['wd']
num_classes = hypes['arch']['num_classes']
vgg_fcn.build(images, train=train, num_classes=num_classes, random_init_fc8=True)
logits = {}
logits['images'] = images
if hypes['arch']['fcn_in'] == 'pool5':
logits['fcn_in'] = vgg_fcn.pool5
elif hypes['arch']['fcn_in'] == 'fc7':
logits['fcn_in'] = vgg_fcn.fc7
else:
raise NotImplementedError
logits['feed2'] = vgg_fcn.pool4
logits['feed4'] = vgg_fcn.pool3
logits['fcn_logits'] = vgg_fcn.upscore32
logits['deep_feat'] = vgg_fcn.pool5
logits['early_feat'] = vgg_fcn.conv4_3
# List what variables to save and restore for finetuning
"""
vars_to_save = {"conv1_1": vgg_fcn.conv1_1, "conv1_2": vgg_fcn.conv1_2,
"pool1": vgg_fcn.pool1,
"conv2_1": vgg_fcn.conv2_1, "conv2_2": vgg_fcn.conv2_2,
"pool2": vgg_fcn.pool2,
"conv3_1": vgg_fcn.conv3_1, "conv3_2": vgg_fcn.conv3_2,
"conv3_3": vgg_fcn.conv3_3, "pool3": vgg_fcn.pool3,
"conv4_1": vgg_fcn.conv4_1, "conv4_2": vgg_fcn.conv4_2,
"conv4_3": vgg_fcn.conv4_3, "pool4": vgg_fcn.pool4,
"conv5_1": vgg_fcn.conv5_1, "conv5_2": vgg_fcn.conv5_2,
"conv5_3": vgg_fcn.conv5_3, "pool5": vgg_fcn.pool5,
"fc6": vgg_fcn.fc6, "fc7": vgg_fcn.fc7}
"""
vars_to_save = (vgg_fcn.conv1_1, vgg_fcn.conv1_2,
vgg_fcn.conv2_1, vgg_fcn.conv2_2,
vgg_fcn.conv3_1, vgg_fcn.conv3_2, vgg_fcn.conv3_3,
vgg_fcn.conv4_1, vgg_fcn.conv4_2, vgg_fcn.conv4_3,
vgg_fcn.conv5_1, vgg_fcn.conv5_2, vgg_fcn.conv5_3,
vgg_fcn.fc6, vgg_fcn.fc7)
logits['saving_vars'] = vars_to_save
return logits
import tensorflow as tf
import os
def inference(hypes, images, train=True):
#Build the MNIST model up to where it may be used for inference.
#Args:
# images: Images placeholder, from inputs().
# train: whether the network is used for train of inference
#Returns:
# softmax_linear: Output tensor with the computed logits.
#vgg16
vgg16_npy_path = os.path.join(hypes['dirs']['data_dir'], "vgg16.npy")
vgg_fcn = fcn8_vgg.FCN8VGG(vgg16_npy_path=vgg16_npy_path)
num_classes = 2
vgg_fcn.build(images, train=train, num_classes=num_classes,
random_init_fc8=True)
vgg_dict = {'unpooled': vgg_fcn.conv5_3,
'deep_feat': vgg_fcn.pool5,
'deep_feat_channels': 512,
'early_feat': vgg_fcn.conv4_3}
return vgg_dict
