def merge_all_classes_to_dataset(csv_files_list, balance = True):
count_direction = []
item_data_list = {}
item_data_selected = {}
for i in range(len(csv_files_list)):
csv_file_path = csv_files_list[i]
temp_data = c_io.read_csv(csv_file_path)
items_data = temp_data[1:] # ignore the first row with labels
count_direction.append(len(items_data))
item_data_list[csv_files_list[i]] = items_data
if balance:
min_count = min(count_direction)
print("old min", min_count)
min_count = min(min_count, 100000)
print("capped min", min_count)
number_per_class = [min(x, min_count) for x in count_direction]
else:
number_per_class = count_direction
print("number for each class ",number_per_class )
for i in range(len(csv_files_list)):
indices = lt.shuffle_indices(count_direction[i], shuffle = True)
indices = indices[:number_per_class[i]]
items_selected = lt.get_list_items_with_indices(item_data_list[csv_files_list[i]], indices)
item_data_selected[csv_files_list[i]] = items_selected
return item_data_selected
def merge_all_scenes_to_classes(src_scenes):
items_for_all_scenes_list = []
csv_file_path_list = []
direction_count = []
for i in range(len(label_list)):
items_for_all_scenes = []
items_for_all_scenes_list.append(items_for_all_scenes)
for scene in src_scenes:
for i in range(len(label_list)):
csv_file_path = os.path.join(
dst_dataset_folder, "{}_{}.csv".format(scene, label_list[i]))
if os.path.exists(csv_file_path):
temp_data = c_io.read_csv(csv_file_path)
items_for_all_scenes_list[
i] = items_for_all_scenes_list[i] + temp_data[
1:] # concatenate two lists together, each list remove the colume names
for i in range(len(label_list)):
csv_file_path = os.path.join(dst_dataset_folder,
"{}.csv".format(label_list[i]))
csv_file_path_list.append(csv_file_path)
direction_count.append(len(items_for_all_scenes_list[i]))
csv_lines = prepare_csv_data(items_for_all_scenes_list[i])
c_io.save_csv(csv_file_path, csv_lines)
return csv_file_path_list, direction_count
def __init__(self, **kwargs):
h5_file = kwargs.pop('h5_file')
csv_file = kwargs.pop('csv_file')
transform_ = kwargs.pop('transform')
option = kwargs.pop('option')
self.data_h5 = h5_file
self.gt_info = c_io.read_csv(csv_file)
if option == "2DScaled" or option == "CuboidTwoStep2DScaled":
print("Reading data from {}".format(h5_file))
self.dataset = cuboidTwoStep2DScaledDatasetH5(
self.data_h5, self.gt_info, transform_)
elif option == "2D" or option == "CuboidTwoStep2D":
self.dataset = cuboidTwoStep2DatasetH5(self.data_h5, self.gt_info,
transform_)
elif option == "5D" or option == "CuboidTwoStep5D":
self.dataset = cuboidTwoStep5DatasetH5(self.data_h5, self.gt_info,
transform_)
elif option == "4D" or option == "CuboidTwoStep4D":
self.dataset = cuboidTwoStep4DatasetH5(self.data_h5, self.gt_info,
transform_)
elif option == "UtilityOnly" or option == "CuboidTwoStepUtilityOnly":
self.dataset = cuboidTwoStepUtilityOnlyDatasetH5(
self.data_h5, self.gt_info, transform_)
super(TwoStepH5Loader, self).__init__(self.dataset, **kwargs)
def __init__(self, csv_file, size, transform=None):
"""
Args:
csv_file (string): Path to the csv file with annotations.
transform (callable, optional): Optional transform to be applied
on a sample.
"""
self.gt_info = c_io.read_csv(csv_file)
self.transform = transform
self.size = size
def __init__(self, csv_file, h, w, transform=None):
"""
Args:
csv_file (string): Path to the csv file with annotations.
transform (callable, optional): Optional transform to be applied
on a sample.
"""
self.gt_info = c_io.read_csv(csv_file)
self.transform = transform
self.height = h
self.width = w
self.label_list = ["up", "down", "left", "right"]
