"--config_filename",
type=str,
default="config.pickle",
help=
"Location to read the metadata related to the training (generated when training)."
)
# parse and validate parameters
args = parser.parse_args()
for k, v in args._get_kwargs():
if isinstance(v, str):
setattr(args, k, v.strip().lower())
# print arguments
rootLogger.info("Running with the following parameters:")
rootLogger.info(vars(args))
# which architecture to use {VGG16, Resnet50}
if args.architecture == 'vgg':
from keras_frcnn.networks import vgg as nn
elif args.architecture == 'resnet':
from keras_frcnn.networks import resnet as nn
else:
rootLogger.info('Not a valid model')
raise ValueError
if not args.model_weight_path: # if filename is not given
rootLogger.info(
'Error: path to test data must be specified. Pass --model_weight_path to command line'
)
def get_anchor_gt(img, C, img_length_calc_function, mode='train'):
"""
This function returns Ground Truth Anchor
:param img:
:param C:
:param img_length_calc_function:
:param mode:
:return:
"""
# Randomly shuffle training data
# if mode == 'train':
# np.random.shuffle(img_data)
# for img in img_data:
try:
# read in image, and optionally add augmentation
# Augment the data in case of training
# if mode == 'train':
# # Returns augmented image and corresponding bboxes
# img_data_aug, x_img = data_augment.augment(img, C, augment=True)
# else:
img_data_aug, x_img = data_augment.augment(img, C, augment=False)
(width, height) = (img_data_aug['width'], img_data_aug['height'])
(rows, cols, _) = x_img.shape
# Cross validate if the augmented data is correct
assert cols == width
assert rows == height
# get image dimensions for resizing
(resized_width,
resized_height) = get_new_img_size(width, height, C.im_size)
# resize the image so that smalles side is length = 600px
x_img = cv2.resize(x_img, (resized_width, resized_height),
interpolation=cv2.INTER_CUBIC)
# TODO : Calculates RPN
try:
y_rpn_cls, y_rpn_regr = calc_rpn(C, img_data_aug, width, height,
resized_width, resized_height,
img_length_calc_function)
except:
rootLogger.info("Exception occured in RPN")
# continue
# Zero-center by mean pixel, and preprocess image
x_img = x_img[:, :, (2, 1, 0)] # BGR -> RGB
x_img = x_img.astype(np.float32)
x_img[:, :, 0] -= C.img_channel_mean[0]
x_img[:, :, 1] -= C.img_channel_mean[1]
x_img[:, :, 2] -= C.img_channel_mean[2]
x_img /= C.img_scaling_factor
x_img = np.transpose(x_img, (2, 0, 1))
x_img = np.expand_dims(x_img, axis=0)
y_rpn_regr[:, y_rpn_regr.shape[1] // 2:, :, :] *= C.std_scaling
x_img = np.transpose(x_img, (0, 2, 3, 1))
y_rpn_cls = np.transpose(y_rpn_cls, (0, 2, 3, 1))
y_rpn_regr = np.transpose(y_rpn_regr, (0, 2, 3, 1))
return np.copy(x_img), [np.copy(y_rpn_cls),
np.copy(y_rpn_regr)], img_data_aug
except Exception as e:
rootLogger.info("Exception occured in `anchor_gt`")
def get_data(input_path):
classes_count = {}
class_mapping = {}
visualise = False
# Get csv file for data
train_annot_path = os.path.join(input_path, 'train.csv')
val_annot_path = os.path.join(input_path, 'train.csv')
# Get Image Path
imgsets_path_train = os.path.join(input_path, 'train')
imgsets_path_val = os.path.join(input_path, 'train')
rootLogger.info('Parsing annotation files')
train_files = []
val_files = []
train_annotation_data = []
val_annotation_data = []
try:
# Training Data
with open(train_annot_path) as f:
csv_reader = csv.reader(f, delimiter=',')
line_count = 0
for row in csv_reader:
if line_count == 0:
line_count += 1
elif np.abs(int(row[4]) - int(row[6])) > 50 and np.abs(
int(row[5]) - int(row[7])) > 50:
train_files.append(row[0])
print(row[0])
annotation_data = {
'filepath':
os.path.join(imgsets_path_train, row[0]),
'width':
row[1],
'height':
row[2],
'bboxes': [{
'class': row[3],
'x1': int(row[4]),
'x2': int(row[6]),
'y1': int(row[5]),
'y2': int(row[7])
}]
}
train_annotation_data.append(annotation_data)
class_name = annotation_data['bboxes'][0]['class']
if class_name not in classes_count:
classes_count[class_name] = 1
break
else:
classes_count[class_name] += 1
if class_name not in class_mapping:
class_mapping[class_name] = len(class_mapping)
except Exception as e:
print(e)
rootLogger.info("Exception in loading training data")
try:
# Validation Data
with open(val_annot_path) as f:
csv_reader = csv.reader(f, delimiter=',')
line_count = 0
for row in csv_reader:
if line_count == 0:
line_count += 1
else:
val_files.append(row[0])
print(row[0])
annotation_data = {
'filepath':
os.path.join(imgsets_path_val, row[0]),
'width':
row[1],
'height':
row[2],
'bboxes': [{
'class': row[3],
'x1': int(row[4]),
'x2': int(row[6]),
'y1': int(row[5]),
'y2': int(row[7])
}]
}
val_annotation_data.append(annotation_data)
class_name = annotation_data['bboxes'][0]['class']
if class_name not in classes_count:
classes_count[class_name] = 1
else:
classes_count[class_name] += 1
break
if class_name not in class_mapping:
class_mapping[class_name] = len(class_mapping)
except Exception as e:
print(e)
rootLogger.info("Exception in loading Validation data")
# all_imgs = train_annotation_data + val_annotation_data
return train_annotation_data, val_annotation_data, classes_count, class_mapping
default="config.pickle",
help=
"Location to store all the metadata related to the training (to be used when testing)."
)
# parse and validate parameters
args = parser.parse_args()
for k, v in args._get_kwargs():
if isinstance(v, str):
setattr(args, k, v.strip().lower())
os.environ['CUDA_VISIBLE_DEVICES'] = str(args.gpu) if args.gpu > -1 else '-1'
# print arguments
rootLogger.info("Running with the following parameters:")
rootLogger.info(vars(args))
# pass the settings from the command line, and persist them in the config object
C = config.Config()
# Use this augmentation only in case of training data, not the validation data
C.use_horizontal_flips = bool(args.horizontal_flips)
C.use_vertical_flips = bool(args.vertical_flips)
C.rot_90 = bool(args.rot_90)
C.model_path = args.output_weight_path
C.num_rois = int(args.num_rois)
# which architecture to use {VGG16, Resnet50}
if args.architecture == 'vgg':