"""
_transform = [
A.Lambda(image=preprocessing_fn),
]
return A.Compose(_transform)
# BACKBONE = 'efficientnetb3'
BACKBONE = args.backbone
BATCH_SIZE = args.batch_size
CLASSES = ['live', 'inter', 'dead']
LR = args.lr
EPOCHS = args.epoch
preprocess_input = sm.get_preprocessing(BACKBONE)
# define network parameters
n_classes = 1 if len(CLASSES) == 1 else (len(CLASSES) + 1) # case for binary and multiclass segmentation
activation = 'sigmoid' if n_classes == 1 else 'softmax'
#create model
net_func = globals()[args.net_type]
encoder_weights='imagenet' if args.pre_train else None
if not args.net_type=='PSPNet':
model = net_func(BACKBONE, encoder_weights=encoder_weights, classes=n_classes, activation=activation,\
decoder_block_type = args.upsample,\
decoder_filters=(int(args.filters),int(args.filters/2), int(args.filters/4), int(args.filters/8), int(args.filters/16)))
print('{}'.format((int(args.filters),int(args.filters/2), int(args.filters/4), int(args.filters/8), int(args.filters/16))))
# network
if not best_flag:
model_file = model_folder + '/best_model-{:03d}.h5'.format(epoch)
print(os.path.exists(model_file))
if os.path.exists(model_file):
best_weight = model_file
else:
best_weight = model_folder + '/best_model.h5'
else:
best_weight = model_folder + '/weights_{:02d}.h5'.format(
epoch) if epoch < 100 else model_folder + '/weights_{:03d}.h5'.format(
epoch)
preprocess_input = sm.get_preprocessing(backbone)
#create model
n_classes = cho if not fl_ch == 'fl12' else 2
activation = 'relu'
net_func = globals()[net_arch]
model = net_func(backbone,
classes=n_classes,
encoder_weights=None,
activation=activation)
#load best weights
model.load_weights(best_weight)
## save model
model.save(model_folder + '/ready_model.h5')
