59d84cd18324aa86373091c8a0c262e6536d32fe,src/detection/tensor/detector.py,Detector,detect,#Detector#,29

Before Change

            // _ = tf.import_graph_def(graph_def, name="")

    def detect(self, image):
        image_array = self._pil_to_tf(image)

        with tf.Graph().as_default() as imported_graph:
            tf.import_graph_def(self.graph_def, name="")

        with tf.Session(graph=imported_graph) as sess:
            with tf.device("/gpu:0"):
                softmax_tensor = sess.graph.get_tensor_by_name("final_result:0")
                predictions = sess.run(softmax_tensor, {"DecodeJpeg:0": image_array})
                predictions = np.squeeze(predictions)
                answer = {}
                for node_id in range(len(predictions)):
                    answer[self.labels[node_id]] = predictions[node_id]

After Change

            return graph_def

    def detect(self, images):
        np_images = [self._pil_to_np(image) for image in images]

        pool = ThreadPool()
        with tf.Graph().as_default() as imported_graph:
            tf.import_graph_def(self.graph_def, name="")

        with tf.Session(graph=imported_graph) as sess:
            with tf.device("/gpu:0"):
                softmax_tensor = sess.graph.get_tensor_by_name("final_result:0")
                threads = [pool.apply_async(self.operation,
                                            args=(sess, softmax_tensor, np_images[image_number], image_number,)) for
                           image_number in range(len(np_images))]
                answers = []
                for thread in threads:
                    prediction, image_number = thread.get()
                    prediction = np.squeeze(prediction)
                    answer = {"image_number": image_number}
                    for node_id in range(len(prediction)):
                        answer[self.labels[node_id]] = prediction[node_id]
                    answers.append(answer)
                return answers

    @staticmethod
    def _pil_to_np(image):

Instances