def __init__(self,
data,
dataset,
batch_size,
shape_rgb,
shape_depth,
is_flip=False,
is_addnoise=False,
is_erase=False):
self.data = data
self.dataset = dataset
self.policy = BasicPolicy(color_change_ratio=0.50,
mirror_ratio=0.50,
flip_ratio=0.0 if not is_flip else 0.2,
add_noise_peak=0 if not is_addnoise else 20,
erase_ratio=-1.0 if not is_erase else 0.5)
self.batch_size = batch_size
self.shape_rgb = shape_rgb
self.shape_depth = shape_depth
self.maxDepth = 1000.0
from sklearn.utils import shuffle
self.dataset = shuffle(self.dataset, random_state=0)
self.N = len(self.dataset)
def __init__(self,
root,
dataset,
batch_size,
shape_rgb,
shape_depth,
locations=False,
is_flip=False,
is_addnoise=False,
is_erase=False,
is_skip_policy=False):
self.root = root
self.shape_rgb = [
(min(shape_rgb), max(shape_rgb)),
(max(shape_rgb), min(shape_rgb)),
]
self.shape_depth = [
(min(shape_depth), max(shape_depth)),
(max(shape_depth), min(shape_depth)),
]
self.dataset = [{
datum: pickle.load(
open(os.path.join(dataset, orientation, datum + '_MD.p'),
'rb'))
for datum in ["imgs", "targets"]
} for orientation in ["landscape", "portrait"]]
self.policy = BasicPolicy(color_change_ratio=0.50,
mirror_ratio=0.50,
flip_ratio=0.0 if not is_flip else 0.2,
add_noise_peak=0 if not is_addnoise else 20,
erase_ratio=-1.0 if not is_erase else 0.5)
self.batch_size = batch_size
self.N = [len(dataset["imgs"]) for dataset in self.dataset]
self.is_skip_policy = is_skip_policy
self.locations = locations
def __init__(self,
data,
dataset,
batch_size,
shape_rgb,
shape_output,
is_flip=False,
is_addnoise=False,
is_erase=False):
self.data = data
self.dataset = dataset
self.policy = BasicPolicy(color_change_ratio=0.50,
mirror_ratio=0.50,
flip_ratio=0.0 if not is_flip else 0.2,
add_noise_peak=0 if not is_addnoise else 20,
erase_ratio=-1.0 if not is_erase else 0.5)
self.batch_size = batch_size
self.shape_rgb = shape_rgb
self.shape_output = shape_output
#Shuffling dataset as continous frames can lead to overfitting
from sklearn.utils import shuffle
self.dataset = shuffle(self.dataset, random_state=0)
self.N = len(self.dataset)
def __init__(self,
data_root,
data_paths,
batch_size,
shape_depth,
channels=5,
is_flip=False,
is_addnoise=False,
is_erase=False,
dont_interpolate=False):
self.data_root = data_root
self.dataset = data_paths
self.policy = BasicPolicy(color_change_ratio=0.50,
mirror_ratio=0.50,
flip_ratio=0.0 if not is_flip else 0.2,
add_noise_peak=0 if not is_addnoise else 20,
erase_ratio=-1.0 if not is_erase else 0.5)
self.batch_size = batch_size
self.channels = channels
self.shape_rgbd = (batch_size, 480, 640, self.channels)
self.shape_depth = shape_depth
self.minDepth = settings.MIN_DEPTH * settings.DEPTH_SCALE #cm
self.maxDepth = settings.MAX_DEPTH * settings.DEPTH_SCALE #cm
self.dont_interpolate = dont_interpolate
from sklearn.utils import shuffle
self.dataset = shuffle(self.dataset, random_state=0)
self.N = len(self.dataset)
def __init__(self, data, dataset, batch_size, shape_rgb, shape_depth, is_flip=False, is_addnoise=False, is_erase=False, is_skip_policy=False):
self.data = data
self.dataset = dataset
self.policy = BasicPolicy( color_change_ratio=0.50, mirror_ratio=0.50, flip_ratio=0.0 if not is_flip else 0.2,
add_noise_peak=0 if not is_addnoise else 20, erase_ratio=-1.0 if not is_erase else 0.5)
self.batch_size = batch_size
self.shape_rgb = shape_rgb
self.shape_depth = shape_depth
self.maxDepth = 1000.0
self.N = len(self.dataset)
self.is_skip_policy = is_skip_policy
def __init__(self,
data_path,
min_depth,
max_depth,
batch_size,
image_shape,
depth_shape,
n_channels,
is_augment=False,
shuffle=False,
mode="train",
is_flip=False,
is_addnoise=False,
is_erase=False,
is_scale=False,
is_deconv=False,
select_label_mode=None,
scale_value=None):
self.data_path = data_path
self.batch_size = batch_size
self.image_shape = image_shape
self.depth_shape = depth_shape
self.n_channels = n_channels
self.shuffle = shuffle
self.is_augment = is_augment
self.min_depth = min_depth
self.max_depth = max_depth
self.is_deconv = is_deconv
self.scale_value = scale_value
self.select_label_mode = select_label_mode
self.policy = BasicPolicy(image_shape=image_shape,
color_change_ratio=0.50,
mirror_ratio=0.50,
flip_ratio=0.0 if not is_flip else 0.2,
add_noise_peak=0 if not is_addnoise else 20,
erase_ratio=-1.0 if not is_erase else 0.5,
scale_ratio=-1 if not is_scale else 0.2)
train_csv_path = os.path.join(self.data_path, "data/nyu2_train.csv")
valid_csv_path = os.path.join(self.data_path, "data/nyu2_test.csv")
if mode == "train":
self.imagePath_labelPath_list = self.read_csv(train_csv_path)
np.random.shuffle(self.imagePath_labelPath_list)
self.on_epoch_end()
else:
self.imagePath_labelPath_list = self.read_csv(valid_csv_path)