parser.add_argument('--batch_size',
help='The training batch size.',
type=int,
default=5)
parser.add_argument('--weights',
help='The path of weights to be loaded.',
type=str,
default=None)
args = parser.parse_args()
# check related paths
paths = check_related_path(os.getcwd())
# get image and label file names for training and validation
_, _, _, _, test_image_names, test_label_names = get_dataset_info(args.dataset)
# build the model
net, base_model = builder(args.num_classes,
(args.crop_height, args.crop_width), args.model,
args.base_model)
# summary
net.summary()
# load weights
print('Loading the weights...')
if args.weights is None:
net.load_weights(filepath=os.path.join(
paths['weigths_path'], '{model}_based_on_{base_model}.h5'.format(
model=args.model, base_model=base_model)))
parser.add_argument('--data_aug_rate', help='The rate of data augmentation.', type=float, default=0.)
parser.add_argument('--checkpoint_freq', help='How often to save a checkpoint.', type=int, default=5)
parser.add_argument('--validation_freq', help='How often to perform validation.', type=int, default=1)
parser.add_argument('--num_valid_images', help='The number of images used for validation.', type=int, default=20)
parser.add_argument('--data_shuffle', help='Whether to shuffle the data.', type=str2bool, default=True)
parser.add_argument('--random_seed', help='The random shuffle seed.', type=int, default=None)
parser.add_argument('--weights', help='The path of weights to be loaded.', type=str, default=None)
parser.add_argument('--steps_per_epoch', help='The training steps of each epoch', type=int, default=None)
args = parser.parse_args()
# check related paths
paths = check_related_path(os.getcwd())
# get image and label file names for training and validation
train_image_names, train_label_names, valid_image_names, valid_label_names, _, _ = get_dataset_info(args.dataset)
# build the model
net, base_model = builder(args.num_classes, (args.crop_height, args.crop_width), args.model, args.base_model)
# summary
net.summary()
# load weights
if args.weights is not None:
print('Loading the weights...')
net.load_weights(args.weights)
# chose loss
losses = {'CE': categorical_crossentropy_with_logits,
'Focal_Loss': focal_loss()}