TensorFlow Lite Object Detection on PC(CentOS7)

Object Detection API と同じディレクトリーで作成して実行します。
models/research/object_detection/

取り敢えず、プログラムはこちらです。

object_detection_lite_cam.py
# -*- coding: utf-8 -*- from __future__ import absolute_import from __future__ import division from __future__ import print_function ''' object_detection_lite_cam.py https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/running_on_mobile_tensorflowlite.md https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/lite/toco/g3doc/python_api.md Created on 2018/09/04 @author: nishi TensorFlow Lite Python interpreter Using the interpreter from a model file ''' import os import numpy as np import tensorflow as tf from PIL import Image import cv2 import sys from object_detection.utils import label_map_util from object_detection.utils import visualization_utils as vis_util BASE_DIR='object_detection' floating_model = False def load_labels(filename): my_labels = [] input_file = open(filename, 'r') for l in input_file: my_labels.append(l.strip()) return my_labels # List of the strings that is used to add correct label for each box. PATH_TO_LABELS = os.path.join(BASE_DIR,'data', 'mscoco_label_map.pbtxt') NUM_CLASSES = 90 print('PATH_TO_LABELS=',PATH_TO_LABELS) label_map = label_map_util.load_labelmap(PATH_TO_LABELS) categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True) category_index = label_map_util.create_category_index(categories) #print('category_index=',category_index) # Load TFLite model and allocate tensors. # /home/nishi/tmp/mobilenet_v1_1.0_224.tflite # /home/nishi/tmp/mobilenet_v1_1.0_224_quant.tflite #interpreter = tf.contrib.lite.Interpreter(model_path="converted_model.tflite") #TF_MODEL='/home/nishi/tmp/mobilenet_v1_1.0_224/mobilenet_v1_1.0_224.tflite' TF_MODEL='/home/nishi/tmp/tflite/detect.tflite' interpreter = tf.contrib.lite.Interpreter(model_path=TF_MODEL) interpreter.allocate_tensors() #img_path="/home/nishi/tmp/mobilenet_v1_1.0_224/grace_hopper.bmp" #label_file="/home/nishi/tmp/mobilenet_v1_1.0_224/labels.txt" input_mean=127.5 input_std=127.5 # Get input and output tensors. input_details = interpreter.get_input_details() output_details = interpreter.get_output_details() # check the type of the input tensor if input_details[0]['dtype'] == np.float32: floating_model = True # NxHxWxC, H:1, W:2 height = input_details[0]['shape'][1] # 入力の高さ width = input_details[0]['shape'][2] # 入力の幅 print("height=",height) print("width=",width) # OK #URL="http://200.36.58.250/mjpg/video.mjpg?resolution=320x240" URL="http://200.36.58.250/mjpg/video.mjpg?resolution=640x480" ''' Web camera ''' if True: video = cv2.VideoCapture(URL) # Exit if video not opened. if not video.isOpened(): print("Could not open video") sys.exit() ''' USBカメラオープン ''' if False: video = cv2.VideoCapture(0) # Exit if video not opened. if not video.isOpened(): print("Could not open video") sys.exit() #video.set(3,width) # 画像サイズの指定 #video.set(4,height) # Test model on random input data. input_shape = input_details[0]['shape'] #labels = load_labels(label_file) while True: #img = Image.open(img_path) #img = img.resize((width, height)) # Read first frame. ok, image_np = video.read() if not ok: print('Cannot read video file') sys.exit() #print("image_np.shape=",image_np.shape) image_np_x = cv2.resize(image_np, (height,width)) # Expand dimensions since the model expects images to have shape: [1, None, None, 3] input_data = np.expand_dims(image_np_x, axis=0) #image_np_expanded = image_np_expanded.npresize((width, height)) # add N dim #input_data = np.expand_dims(img, axis=0) #print("input_data.shape=",input_data.shape) if floating_model: #print('passed :#3') input_data = (np.float32(input_data) - input_mean) / input_std #input_data = np.array(np.random.random_sample(input_shape), dtype=np.float32) interpreter.set_tensor(input_details[0]['index'], input_data) interpreter.invoke() #boxes, scores, classes, num boxes = interpreter.get_tensor(output_details[0]['index']) classes = interpreter.get_tensor(output_details[1]['index']) scores = interpreter.get_tensor(output_details[2]['index']) num = int(interpreter.get_tensor(output_details[3]['index'])[0]) if(num > 0): print('------------') print('num',num) #print('boxes=',boxes) print('classes=',classes[0]) print('scores=',scores[0]) #top_k = results.argsort()[-4:][::-1] #labels = load_labels(label_file) for i in range(num): classes[0][i]=classes[0][i] + 1.0 id = int(classes[0][i]) sco=scores[0][i] name='none' if(id in category_index): s=category_index[id] name=s['name'] print(name,'/',sco) # Visualization of the results of a detection. vis_util.visualize_boxes_and_labels_on_image_array( image_np, np.squeeze(boxes), np.squeeze(classes).astype(np.int32), np.squeeze(scores), category_index, use_normalized_coordinates=True, min_score_thresh=0.3, # add by nishi max_boxes_to_draw=num, # add by nishi line_thickness=2) else: print('num',num) #plt.figure(figsize=IMAGE_SIZE) #plt.imshow(image_np) #plt.show() cv2.imshow('img',image_np) #---------------- # プログラムの終了は、 q を入力してください。 #---------------- if cv2.waitKey(1) & 0xFF == ord('q'): #if cv2.waitKey(5) & 0xFF == ord('q'): #if cv2.waitKey(3) & 0xFF == ord('q'): break video.release() cv2.destroyAllWindows()

起動方法は、
$export PYTHONPATH=xxx/models/research:xxx/models/research/slim:$PYTHONPATH $cd xxx/models/research $python3.5 object_detection/object_detection_lite_cam.py