def feeder_factory(file_list,
reader_type,
feeder_type="npyfile",
stereo=False):
if feeder_type == "npyfile":
if stereo and reader_type != "extrinsic":
FeederClass = NpyFileFeederStereoLeft
else:
FeederClass = NpyFileFeeder
else:
raise WrongInputException("Wrong feeder type: " + feeder_type)
if reader_type == "image":
reader = ImageReaderStereo() if stereo else ImageReader()
elif reader_type == "intrinsic":
reader = NpyTxtReaderStereo() if stereo else NpyTxtReader()
elif reader_type == "depth":
reader = DepthReaderStereo() if stereo else DepthReader()
elif reader_type == "pose":
reader = PoseReaderStereo() if stereo else PoseReader()
elif reader_type == "extrinsic":
reader = NpyTxtReader()
else:
raise WrongInputException("Wrong reader type: " + reader_type)
feeder = FeederClass(file_list, reader)
return feeder
def optimizer_factory(opt_name, basic_lr, epoch=0):
if opt_name == "adam_constant":
return tf.optimizers.Adam(learning_rate=basic_lr)
if opt_name == "sgd_constant":
return tf.optimizers.SGD(learning_rate=basic_lr)
else:
raise WrongInputException(f"{opt_name} is NOT an available model name")
def list_drive_paths(self):
split_path = op.join(self.base_path,
self.left_img_dir + self.dir_suffix, self.split)
if not op.isdir(split_path):
raise WrongInputException(
"[list_sequence_paths] path does NOT exist:" + split_path)
city_names = os.listdir(split_path)
city_names = [
city for city in city_names if op.isdir(op.join(split_path, city))
]
total_sequences = []
for city in city_names:
pattern = op.join(split_path, city, "*.png")
files = glob(pattern)
seq_numbers = [file.split("_")[-3] for file in files]
seq_numbers = list(set(seq_numbers))
seq_numbers.sort()
seq_paths = [
op.join(self.split, city, f"{city}_{seq}")
for seq in seq_numbers
]
total_sequences.extend(seq_paths)
# total_sequences: list of ["city/city_seqind"]
print("[list_drive_paths]", total_sequences)
return total_sequences
def flow_net_factory(self, net_name, conv2d_f):
if net_name == "PWCNet":
flownet = PWCNet(self.input_shape, conv2d_f)()
else:
raise WrongInputException(
"[flow_net_factory] wrong flow net name: " + net_name)
return flownet
def pose_net_factory(self, net_name, conv2d_p):
if net_name == "PoseNet":
posenet = PoseNet(self.input_shape, conv2d_p)()
else:
raise WrongInputException(
"[pose_net_factory] wrong pose net name: " + net_name)
return posenet
def __init__(self, method):
if method == "L1":
self.photometric_loss = photometric_loss_l1
elif method == "SSIM":
self.photometric_loss = photometric_loss_ssim
else:
raise WrongInputException("Wrong photometric loss name: " + method)
def activation_factory(self, activ_name):
if activ_name == "InverseSigmoid":
return InverseSigmoidActivation()
elif activ_name == "Exponential":
return ExponentialActivation()
else:
WrongInputException(
"[activation_factory] wrong activation name: " + activ_name)
def set_type_and_shape(self, value):
if isinstance(value, np.ndarray):
if value.dtype == np.uint8:
self.decode_type = "tf.uint8"
elif value.dtype == np.float32:
self.decode_type = "tf.float32"
else:
raise WrongInputException(
f"[FeederBase] Wrong numpy type: {value.dtype}")
self.parse_type = "tf.string"
self.shape = list(value.shape)
elif isinstance(value, int):
self.parse_type = "tf.int64"
self.shape = None
else:
raise WrongInputException(
f"[FeederBase] Wrong type: {type(value)}")
def depth_net_factory(self, net_name, conv2d_d, pred_activ,
upsample_interp):
if net_name == "DepthNetBasic":
depth_net = DepthNetBasic(self.input_shape, conv2d_d, pred_activ,
upsample_interp)()
elif net_name == "DepthNetNoResize":
depth_net = DepthNetNoResize(self.input_shape, conv2d_d,
pred_activ, upsample_interp)()
elif net_name in PRETRAINED_MODELS:
depth_net = DepthNetFromPretrained(self.input_shape, conv2d_d,
pred_activ, upsample_interp,
net_name,
self.pretrained_weight)()
else:
raise WrongInputException(
"[depth_net_factory] wrong depth net name: " + net_name)
return depth_net
def dataset_reader_factory(raw_data_path, dataset, split, stereo=opts.STEREO):
if dataset == "kitti_raw" and split == "train":
data_reader = kr.KittiRawTrainReader(raw_data_path, stereo)
elif dataset == "kitti_raw" and split == "test":
data_reader = kr.KittiRawTestReader(raw_data_path, stereo)
elif dataset == "kitti_odom" and split == "train":
data_reader = kr.KittiOdomTrainReader(raw_data_path, stereo)
elif dataset == "kitti_odom" and split == "test":
data_reader = kr.KittiOdomTestReader(raw_data_path, stereo)
elif dataset == "cityscapes":
data_reader = cr.CityScapesReader(raw_data_path, stereo, split)
elif dataset == "cityscapes_seq":
data_reader = cr.CityScapesReader(raw_data_path, stereo, split,
"_sequence")
else:
raise WrongInputException(
f"Wrong dataset and split: {dataset}, {split}")
return data_reader
def encode(self, input_image):
"""
:param input_image: (batch, height, width, channel)
"""
input_shape = input_image.get_shape()
height = input_shape[1]
net_name = self.net_name
weights = "imagenet" if self.pretrained_weight else None
jsonfile = op.join(opts.PROJECT_ROOT, "model", "build_model",
"scaled_layers.json")
output_layers = self.read_output_layers(jsonfile)
out_layer_names = output_layers[net_name]
if net_name == "MobileNetV2":
from tensorflow.keras.applications.mobilenet_v2 import preprocess_input
pproc_img = layers.Lambda(lambda x: preprocess_input(x),
name="preprocess_mobilenet")(input_image)
ptmodel = tfapp.MobileNetV2(input_shape=input_shape,
include_top=False,
weights=weights)
elif net_name == "NASNetMobile":
from tensorflow.keras.applications.nasnet import preprocess_input
assert height == 128
def preprocess_layer(x):
x = preprocess_input(x)
x = tf.image.resize(x,
size=NASNET_SHAPE[:2],
method="bilinear")
return x
pproc_img = layers.Lambda(lambda x: preprocess_layer(x),
name="preprocess_nasnet")(input_image)
ptmodel = tfapp.NASNetMobile(input_shape=NASNET_SHAPE,
include_top=False,
weights=weights)
elif net_name == "DenseNet121":
from tensorflow.keras.applications.densenet import preprocess_input
pproc_img = layers.Lambda(lambda x: preprocess_input(x),
name="preprocess_densenet")(input_image)
ptmodel = tfapp.DenseNet121(input_shape=input_shape,
include_top=False,
weights=weights)
elif net_name == "VGG16":
from tensorflow.keras.applications.vgg16 import preprocess_input
pproc_img = layers.Lambda(lambda x: preprocess_input(x),
name="preprocess_vgg16")(input_image)
ptmodel = tfapp.VGG16(input_shape=input_shape,
include_top=False,
weights=weights)
elif net_name == "Xception":
from tensorflow.keras.applications.xception import preprocess_input
assert height == 128
def preprocess_layer(x):
x = preprocess_input(x)
x = tf.image.resize(x,
size=XCEPTION_SHAPE[:2],
method="bilinear")
return x
pproc_img = layers.Lambda(lambda x: preprocess_layer(x),
name="preprocess_xception")(input_image)
ptmodel = tfapp.Xception(input_shape=XCEPTION_SHAPE,
include_top=False,
weights=weights)
elif net_name == "ResNet50V2":
from tensorflow.keras.applications.resnet import preprocess_input
pproc_img = layers.Lambda(lambda x: preprocess_input(x),
name="preprocess_resnet")(input_image)
ptmodel = tfapp.ResNet50V2(input_shape=input_shape,
include_top=False,
weights=weights)
elif net_name == "NASNetLarge":
from tensorflow.keras.applications.nasnet import preprocess_input
assert height == 128
def preprocess_layer(x):
x = preprocess_input(x)
x = tf.image.resize(x,
size=NASNET_SHAPE[:2],
method="bilinear")
return x
pproc_img = layers.Lambda(lambda x: preprocess_layer(x),
name="preprocess_nasnet")(input_image)
ptmodel = tfapp.NASNetLarge(input_shape=NASNET_SHAPE,
include_top=False,
weights=weights)
else:
raise WrongInputException("Wrong pretrained model name: " +
net_name)
# collect multi scale convolutional features
layer_outs = []
for layer_name in out_layer_names:
layer = ptmodel.get_layer(name=layer_name[1], index=layer_name[0])
# print("extract feature layers:", layer.name, layer.get_input_shape_at(0), layer.get_output_shape_at(0))
layer_outs.append(layer.output)
# create model with multi scale features
multi_scale_model = tf.keras.Model(ptmodel.input,
layer_outs,
name=net_name + "_base")
features_ms = multi_scale_model(pproc_img)
return features_ms