loss_name = args.loss
poly_lr = args.poly_learning_rate
lambda_loss = args.lambda_loss
pixel_distance = args.pixel_distance
num_cl = 108
train_sz = 4498
valid_sz = 500
rsize = True
print('Model')
deeplab_model = ResNet101(input_shape_1=(None, None, 3),
input_shape_2=(None, None, 21),
classes=num_cl,
kernel_init=kernel_init,
batch_norm=batchNorm,
dil_rate_model=dil_rate,
mult_rate=mult_rate)
pathLoadWeights = "Y:/tesisti/rossi/Weights/prova.h5"
deeplab_model.load_weights(pathLoadWeights, True)
if loss_name == "standard":
loss = ["categorical_crossentropy"]
loss_name = "standard_loss"
elif loss_name == "custom_loss":
loss = [custom_loss(deeplab_model.get_layer('logits'))]
loss_name = "custom_loss"
assert not (pathCP == 'not_defined'), "Checkpoint path not defined"
if pathSave == 'not_defined':
pathSave = "./"
else:
pathSave = pathSave + "/"
pT = pathSave + datetime.now().strftime("%Y%m%d-%H%M%S")
print(pathCP)
if not os.path.isdir(pT):
os.mkdir(pT)
inf_path = pT + "/inf"
os.mkdir(inf_path)
print("load model weights")
deeplab_model = ResNet101(input_shape_1=(None, None, 3), input_shape_2=(None, None, 21), classes=n_class)
deeplab_model.load_weights(pathCP, by_name=True)
dir_img = "Y:/tesisti/rossi/data/train_val_test_png/test_png/"
dir_softmax = "Y:/tesisti/rossi/data/segmentation_gray/results_softmax_deeplabV2/test/"
dir_seg = "Y:/tesisti/rossi/data/segmentation_part_gray/new_dataset_107/data_part_107part_test/"
####
images = glob.glob(dir_img + "*.png")
images.sort()
softmax = glob.glob(dir_softmax + "*.npy")
softmax.sort()
segs = glob.glob(dir_seg + "*.png")
segs.sort()