def predict_from_h5data(self, h5_data_path, val_fold_nb, use_channels, is_train=False, save_dir=None,
random_k_fold=False, random_k_fold_npy=None, input_channels=1,
output_channels=2, random_crop_size=None, mask_nb=0, batch_size=4
):
dataset = DataSet(h5_data_path, val_fold_nb, random_k_fold=random_k_fold, random_k_fold_npy=random_k_fold_npy,
input_channels=input_channels,
output_channels=output_channels, random_crop_size=random_crop_size, mask_nb=mask_nb,
batch_size=batch_size)
images, _ = dataset.prepare_data(is_train)
pred = self.predict(images, batch_size, use_channels=use_channels)
if save_dir:
keys = dataset.get_keys(is_train)
mask_file_lst = ["{:03}.tif".format(int(key)) for key in keys]
self.save_mask(pred, mask_file_lst, mask_nb, result_save_dir=save_dir)
return pred
def do_evaluate():
h5_data_path = "/home/topsky/helloworld/study/njai_challenge/cbct/inputs/data_0717.hdf5"
val_fold_nb = "01"
output_channels = 2
is_train = False
model_def = get_se_inception_resnet_v2_unet_sigmoid_gn(weights=None, output_channels=output_channels)
model_weights = "/home/topsky/helloworld/study/njai_challenge/cbct/model_weights/new/20180802_0/se_inception_resnet_v2_gn_fold01_random_kfold_0_1i_2o.h5"
model_obj = ModelDeployment(model_def, model_weights)
dataset = DataSet(h5_data_path, val_fold_nb=val_fold_nb)
images, masks = dataset.prepare_data(is_train=is_train)
print(images.shape)
print(masks.shape)
# idx_lst = [0, 5, 10, 15, 20]
# val_images = np.array([images[i] for i in idx_lst])
# val_masks = np.array([masks[i] for i in idx_lst])
# model_obj.evaluate(val_images, val_masks)
model_obj.evaluate(images, masks)