[object detection] 동영상 코드

import tensorflow as tf
		import os
		import pathlib
		import numpy as np
		import zipfile
		import matplotlib.pyplot as plt
		from PIL import Image
		import cv2
		from object_detection.utils import ops as utils_ops
		from object_detection.utils import label_map_util
		from object_detection.utils import visualization_utils as viz_utils
		import time
		# 내 로컬에 설치된 레이블 파일을, 인덱스와 연결시킨다.
		PATH_TO_LABELS = 'C:\\Users\\405\\Documents\\TensorFlow\\models\\research\\object_detection\\data\\mscoco_label_map.pbtxt'
		category_index = label_map_util.create_category_index_from_labelmap(PATH_TO_LABELS)
		# 모델 로드하는 함수.
		# https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/tf2_detection_zoo.md
		# 위의 사이트에서 모델을 가져올수있다.
		# /20200711/efficientdet_d0_coco17_tpu-32.tar.gz
		# Download and extract model
		def download_model(model_name, model_date):
		base_url = 'http://download.tensorflow.org/models/object_detection/tf2/'
		model_file = model_name + '.tar.gz'
		model_dir = tf.keras.utils.get_file(fname=model_name,
		origin=base_url + model_date + '/' + model_file,
		untar=True)
		return str(model_dir)
		MODEL_DATE = '20200711'
		MODEL_NAME = 'centernet_hg104_1024x1024_coco17_tpu-32'
		PATH_TO_MODEL_DIR = download_model(MODEL_NAME, MODEL_DATE)
		def load_model(model_dir) :
		model_full_dir = model_dir + "/saved_model"
		# Load saved model and build the detection function
		detection_model = tf.saved_model.load(model_full_dir)
		return detection_model
		detection_model = load_model(PATH_TO_MODEL_DIR)
		def show_inference(detection_model, image_np) :
		# The input needs to be a tensor, convert it using `tf.convert_to_tensor`.
		input_tensor = tf.convert_to_tensor(image_np)
		# The model expects a batch of images, so add an axis with `tf.newaxis`.
		input_tensor = input_tensor[tf.newaxis, ...]
		# input_tensor = np.expand_dims(image_np, 0)
		detections = detection_model(input_tensor)
		# print(detections)
		# All outputs are batches tensors.
		# Convert to numpy arrays, and take index [0] to remove the batch dimension.
		# We're only interested in the first num_detections.
		num_detections = int(detections.pop('num_detections'))
		detections = {key: value[0, :num_detections].numpy()
		for key, value in detections.items()}
		detections['num_detections'] = num_detections
		# detection_classes should be ints.
		detections['detection_classes'] = detections['detection_classes'].astype(np.int64)
		print(detections)
		image_np_with_detections = image_np.copy()
		viz_utils.visualize_boxes_and_labels_on_image_array(
		image_np_with_detections,
		detections['detection_boxes'],
		detections['detection_classes'],
		detections['detection_scores'],
		category_index,
		use_normalized_coordinates=True,
		max_boxes_to_draw=200,
		min_score_thresh=.30,
		agnostic_mode=False)
		cv2.imshow( 'result' , image_np_with_detections)
		def save_inference(detection_model, image_np, video_writer):
		# The input needs to be a tensor, convert it using `tf.convert_to_tensor`.
		input_tensor = tf.convert_to_tensor(image_np)
		# The model expects a batch of images, so add an axis with `tf.newaxis`.
		input_tensor = input_tensor[tf.newaxis, ...]
		# input_tensor = np.expand_dims(image_np, 0)
		detections = detection_model(input_tensor)
		# print(detections)
		# All outputs are batches tensors.
		# Convert to numpy arrays, and take index [0] to remove the batch dimension.
		# We're only interested in the first num_detections.
		num_detections = int(detections.pop('num_detections'))
		detections = {key: value[0, :num_detections].numpy()
		for key, value in detections.items()}
		detections['num_detections'] = num_detections
		# detection_classes should be ints.
		detections['detection_classes'] = detections['detection_classes'].astype(np.int64)
		print(detections)
		image_np_with_detections = image_np.copy()
		viz_utils.visualize_boxes_and_labels_on_image_array(
		image_np_with_detections,
		detections['detection_boxes'],
		detections['detection_classes'],
		detections['detection_scores'],
		category_index,
		use_normalized_coordinates=True,
		max_boxes_to_draw=200,
		min_score_thresh=.30,
		agnostic_mode=False)
		video_writer.write(image_np_with_detections)
		# 비디오를 실행하는 코드
		cap = cv2.VideoCapture('data/video.mp4')
		if cap.isOpened() == False:
		print('비디오 실행 에러')
		else :
		frame_width = int( cap.get(3) )
		frame_height = int( cap.get(4) )
		out = cv2.VideoWriter('data/output.avi',
		cv2.VideoWriter_fourcc('M','J','P',"G") ,
		20,
		(frame_width,frame_height) )
		# 비디오 캡쳐에서, 이미지를 1장씩 가져온다.
		# 이 1장의 이미지를, 오브젝트 디텍션 한다.
		while cap.isOpened() :
		ret, frame = cap.read()
		if ret == True:
		# frame 이 이미지에 대한 넘파이 어레이 이므로
		# 이 frame 을 오브젝트 디텍션 한다.
		# 학습용으로, 동영상으로 저장하는 코드롤
		# 수정하세요.
		start_time = time.time()
		save_inference(detection_model, frame, out)
		# 동영상을 실시간으로 화면에서 디텍팅하는것
		# show_inference(detection_model, frame)
		end_time = time.time()
		print('연산에 걸린 시간', str(end_time-start_time))
		if cv2.waitKey(27) & 0xFF == 27:
		break
		else :
		break
		cap.release()
		out.release()
		cv2.destroyAllWindows()