def _predict_patches(self, dataset, batch_size):
'''
Using the params.pkl or instantiated model to create patch predictions.
'''
dim = self.dataset_config.output_dim
compute_output = util.create_predictor(dataset, self.model_config, self.params, batch_size)
result_output, result_label = util.batch_predict(compute_output, dataset, dim, batch_size)
return result_output, result_label
def _predict_patches(self, dataset, batch_size):
'''
Using the params.pkl or instantiated model to create patch predictions.
'''
dim = self.dataset_config.output_dim
compute_output = util.create_predictor(dataset, self.model_config,
self.params, batch_size)
result_output, result_label = util.batch_predict(
compute_output, dataset, dim, batch_size)
return result_output, result_label
def visualize(self, image_path, batch_size, best_trade_off=0.1):
dataset, dim = self.create_data_from_image(image_path)
compute_output = util.create_predictor(dataset, self.model_config, self.model_params, batch_size)
predictions, labels = util.batch_predict(compute_output, dataset, self.dim_label, batch_size)
image = self.combine_to_image(predictions, dim)
#Need to have Mass_road structure TODO: argument
dir = os.path.abspath(image_path + "../../../")
#TODO: not the same extension for labels and data. In the case of MASS.
file_ext = os.path.basename(image_path).split('.')[-1]
label_ext = os.listdir(dir + "/labels/")[0].split('.')[-1]
label_path = dir + "/labels/" + os.path.basename(image_path).split('.')[-2] + "." + label_ext
label_image = Image.open(label_path, 'r')
raw_image = Image.open(image_path, 'r')
hit_image = self._create_hit_image(image, raw_image, label_image , best_trade_off)
return image, hit_image, raw_image, label_image
