def load_config():
"""
Loading the config and updating the data arguments(e.g., turn off the use_horizontal_flips, use_vertical_flips and
rot_90) in the config class.
Removing results data in the folder when running this class.
Parsing the data from annotation file for testing process.
"""
curr_config = config.Config()
if not os.path.exists(curr_config.results_image_path):
os.makedirs(curr_config.results_image_path)
if not os.path.exists(curr_config.results_video_path):
os.makedirs(curr_config.results_video_path)
if not os.path.exists(curr_config.result_graphs_path):
os.makedirs(curr_config.result_graphs_path)
if not os.path.exists(curr_config.result_graphs_path):
os.makedirs(curr_config.result_graphs_path)
remove_all_files(curr_config.result_graphs_path)
remove_all_files(curr_config.results_image_path)
remove_all_files(curr_config.results_video_path)
with open(curr_config.train_config_output_filename, 'rb') as f_in:
curr_config = pickle.load(f_in)
# turn off any data augmentation at test time
curr_config.use_horizontal_flips = False
curr_config.use_vertical_flips = False
curr_config.rot_90 = False
if 'bg' not in curr_config.class_mapping:
curr_config.class_mapping['bg'] = len(curr_config.class_mapping)
class_mapping = {v: k for k, v in curr_config.class_mapping.items()}
print('Class Mappings: ')
pprint.pprint(class_mapping)
class_to_color = {class_mapping[v]: np.random.randint(0, 255, 3) for v in class_mapping}
# Save abnormal behaviour classes in class_mapping in config class.
curr_config.class_mapping = class_mapping
gt_records_count, ground_truth_data = get_data(curr_config.train_path_file, header=True, ground_truth=True)
return curr_config, class_to_color, gt_records_count, ground_truth_data
def update_execute_config():
"""
Loading the config and updating the config class.
Removing results data in the folder when running this class.
Setting the path to weights based on Faster R-CNN and VGG16 model.
Parsing the data from annotation file for training process.
:return classes_count: number of abnormal behaviour labels.
:return class_mapping: abnormal behaviour classes(labels).
:return all_imgs: all the training dataset.
:return curr_config: updated config class.
"""
curr_config = config.Config()
print("Clear the graph data in the folder {}".format(
curr_config.result_graphs_path))
remove_file(curr_config.train_process_data_file)
# check if weight path was passed via command line
# Input path for weights. try to load default weights provided by keras.
# set the path to weights based on backend and model
curr_config.base_net_weights = nn.get_weight_path()
# Path to training data.
all_imgs, classes_count, class_mapping = get_data(
curr_config.train_path_file, header=True)
if 'bg' not in classes_count:
classes_count['bg'] = 0
class_mapping['bg'] = len(class_mapping)
curr_config.class_mapping = class_mapping
print('Training images per class: ')
pprint.pprint(classes_count)
print('Num classes (including bg) = {}'.format(len(classes_count)))
with open(curr_config.train_config_output_filename, 'wb') as config_f:
pickle.dump(curr_config, config_f)
print(
'Config has been written to {}, and can be loaded when testing to ensure correct results'
.format(curr_config.train_config_output_filename))
num_imgs = len(all_imgs)
print('Num all images {}'.format(num_imgs))
return classes_count, class_mapping, all_imgs, curr_config
def load_config_train():
"""
Loading the config class and calling the remove_all_files and remove_all_file methods.
Calling the get_data method from fast_rcnn folder to format the temporal_anomaly_annotation.
:return curr_config: modified config class.
:return temporal_anomaly_annotation: modified temporal_anomaly_annotation file.
"""
curr_config = config.Config()
# creating a folder named data
if not os.path.exists(curr_config.train_images_dataset_path):
os.makedirs(curr_config.train_images_dataset_path)
remove_all_files(curr_config.train_images_dataset_path)
remove_file(curr_config.train_path_file)
temporal_anomaly_annotation = get_data(
curr_config.train_videos_annotation_file, temporal_anomaly_data=True)
return curr_config, temporal_anomaly_annotation
from networks import GoogleNet as nn
elif options.net == 'ResNet50':
from networks import ResNet_50 as nn
if K.image_data_format() == 'tf':
base_net_weights = './model/resnet50_weights_th_dim_ordering_th_kernels_notop.h5'
else:
base_net_weights = './model/resnet50_weights_tf_dim_ordering_tf_kernels.h5'
elif options.net == 'ResNet101':
from networks import ResNet_101 as nn
else:
raise ValueError(
"Command line option parser must be one of 'ZF' or 'VGG16' or 'GoogleNet' or 'ResNet50' or 'ResNet101'"
)
# pass the settings from the command line, and persist them in the config object
C = config.Config()
all_imgs, classes_count, class_mapping = get_data(options.train_path)
if 'bg' not in classes_count:
classes_count['bg'] = 0
class_mapping['bg'] = len(class_mapping)
C.class_mapping = class_mapping
inv_map = {v: k for k, v in class_mapping.items()}
print('Training images per class:')
pprint.pprint(classes_count)
print('Num classes (including bg) = {}'.format(len(classes_count)))
config_output_filename = C.config_filename