全文共6821字，預計學習時長20分鐘
從自動駕駛汽車檢測路上的物體，到通過複雜的面部及身體語言識別發現可能的犯罪活動。多年來，研究人員一直在探索讓機器通過視覺識別物體的可能性。
這一特殊領域被稱為計算機視覺 (Computer Vision, CV)，在現代生活中有著廣泛的應用。
目標檢測 (ObjectDetection) 也是計算機視覺最酷的應用之一，這是不容置疑的事實。
現在的CV工具能夠輕鬆地將目標檢測應用於圖片甚至是直播視頻。本文將簡單地展示如何用TensorFlow創建實時目標檢測器。
建立一個簡單的目標檢測器
設置要求：
TensorFlow版本在1.15.0或以上
執行pip install TensorFlow安裝最新版本
一切就緒，現在開始吧！
設置環境
第一步：從Github上下載或複製TensorFlow目標檢測的代碼到本地計算機
在終端運行如下命令：
git clonehttps://github.com/tensorflow/models.git
第二步：安裝依賴項
下一步是確定計算機上配備了運行目標檢測器所需的庫和組件。
下面列舉了本項目所依賴的庫。（大部分依賴都是TensorFlow自帶的）
· Cython
· contextlib2
· pillow
· lxml
· matplotlib
若有遺漏的組件，在運行環境中執行pip install即可。
第三步：安裝Protobuf編譯器
谷歌的Protobuf，又稱Protocol buffers，是一種語言無關、平臺無關、可擴展的序列化結構數據的機制。Protobuf幫助程式設計師定義數據結構，輕鬆地在各種數據流中使用各種語言進行編寫和讀取結構數據。
Protobuf也是本項目的依賴之一。點擊這裡了解更多關於Protobufs的知識。接下來把Protobuf安裝到計算機上。
打開終端或者打開命令提示符，將地址改為複製的代碼倉庫，在終端執行如下命令：
cd models/research \wget -Oprotobuf.zip https://github.com/protocolbuffers/protobuf/releases/download/v3.9.1/protoc-3.9.1-osx-x86_64.zip\unzipprotobuf.zip
注意：請務必在models/research目錄解壓protobuf.zip文件。
第四步：編輯Protobuf編譯器
從research/ directory目錄中執行如下命令編輯Protobuf編譯器：
./bin/protoc object_detection/protos/*.proto--python_out=.
用Python實現目標檢測
現在所有的依賴項都已經安裝完畢，可以用Python實現目標檢測了。
在下載的代碼倉庫中，將目錄更改為：
models/research/object_detection
這個目錄下有一個叫object_detection_tutorial.ipynb的ipython notebook。該文件是演示目標檢測算法的demo，在執行時會用到指定的模型：
ssd_mobilenet_v1_coco_2017_11_17
這一測試會識別代碼庫中提供的兩張測試圖片。下面是測試結果之一：
要檢測直播視頻中的目標還需要一些微調。在同一文件夾中新建一個Jupyter notebook，按照下面的代碼操作：
[1]:
import numpy as npimport osimport six.moves.urllib as urllibimport sysimport tarfileimport tensorflow as tfimport zipfilefrom distutils.version import StrictVersionfrom collections import defaultdictfrom io import StringIOfrom matplotlib import pyplot as pltfrom PIL import Image# This isneeded since the notebook is stored in the object_detection folder.sys.path.append("..")from utils import ops as utils_opsif StrictVersion(tf.__version__) < StrictVersion('1.12.0'): raise ImportError('Please upgrade your TensorFlow installation to v1.12.*.')
[2]:
# This isneeded to display the images.get_ipython().run_line_magic('matplotlib', 'inline')
[3]:
# Objectdetection imports# Here arethe imports from the object detection module.from utils import label_map_utilfrom utils import visualization_utils as vis_util
[4]:
# Modelpreparation# Anymodel exported using the `export_inference_graph.py` tool can be loaded heresimply by changing `PATH_TO_FROZEN_GRAPH` to point to a new .pb file.# Bydefault we use an "SSD with Mobilenet" model here.#See https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md#for alist of other models that can be run out-of-the-box with varying speeds andaccuracies.# Whatmodel to download.MODEL_NAME= 'ssd_mobilenet_v1_coco_2017_11_17'MODEL_FILE= MODEL_NAME + '.tar.gz'DOWNLOAD_BASE= 'http://download.tensorflow.org/models/object_detection/'# Path tofrozen detection graph. This is the actual model that is used for the objectdetection.PATH_TO_FROZEN_GRAPH= MODEL_NAME + '/frozen_inference_graph.pb'# List ofthe strings that is used to add correct label for each box.PATH_TO_LABELS= os.path.join('data', 'mscoco_label_map.pbtxt')
[5]:
#DownloadModelopener =urllib.request.URLopener()opener.retrieve(DOWNLOAD_BASE+ MODEL_FILE, MODEL_FILE)tar_file =tarfile.open(MODEL_FILE)for file in tar_file.getmembers(): file_name= os.path.basename(file.name) if'frozen_inference_graph.pb'in file_name: tar_file.extract(file,os.getcwd())
[6]:
# Load a(frozen) Tensorflow model into memory.detection_graph= tf.Graph()with detection_graph.as_default(): od_graph_def= tf.GraphDef() withtf.gfile.GFile(PATH_TO_FROZEN_GRAPH, 'rb') as fid: serialized_graph= fid.read() od_graph_def.ParseFromString(serialized_graph) tf.import_graph_def(od_graph_def,name='')
[7]:
# Loadinglabel map# Labelmaps map indices to category names, so that when our convolution networkpredicts `5`,#we knowthat this corresponds to `airplane`. Here we use internal utilityfunctions,#butanything that returns a dictionary mapping integers to appropriate stringlabels would be finecategory_index= label_map_util.create_category_index_from_labelmap(PATH_TO_LABELS,use_display_name=True)
[8]:
defrun_inference_for_single_image(image, graph): with graph.as_default(): with tf.Session() as sess: # Get handles to input and output tensors ops= tf.get_default_graph().get_operations() all_tensor_names= {output.name for op in ops for output in op.outputs} tensor_dict= {} for key in [ 'num_detections', 'detection_boxes', 'detection_scores', 'detection_classes', 'detection_masks']: tensor_name= key + ':0' if tensor_name in all_tensor_names: tensor_dict[key]= tf.get_default_graph().get_tensor_by_name(tensor_name) if'detection_masks'in tensor_dict: # The following processing is only for single image detection_boxes= tf.squeeze(tensor_dict['detection_boxes'], [0]) detection_masks= tf.squeeze(tensor_dict['detection_masks'], [0]) # Reframe is required to translate mask from boxcoordinates to image coordinates and fit the image size. real_num_detection= tf.cast(tensor_dict['num_detections'][0], tf.int32) detection_boxes= tf.slice(detection_boxes, [0, 0], [real_num_detection, -1]) detection_masks= tf.slice(detection_masks, [0, 0, 0], [real_num_detection, -1, -1]) detection_masks_reframed= utils_ops.reframe_box_masks_to_image_masks( detection_masks,detection_boxes, image.shape[1],image.shape[2]) detection_masks_reframed= tf.cast( tf.greater(detection_masks_reframed,0.5),tf.uint8) # Follow the convention by adding back the batchdimension tensor_dict['detection_masks'] =tf.expand_dims( detection_masks_reframed,0) image_tensor= tf.get_default_graph().get_tensor_by_name('image_tensor:0') # Run inference output_dict= sess.run(tensor_dict, feed_dict={image_tensor: image}) # all outputs are float32 numpy arrays, so convert typesas appropriate output_dict['num_detections'] =int(output_dict['num_detections'][0]) output_dict['detection_classes'] =output_dict[ 'detection_classes'][0].astype(np.int64) output_dict['detection_boxes'] =output_dict['detection_boxes'][0] output_dict['detection_scores'] =output_dict['detection_scores'][0] if'detection_masks'in output_dict: output_dict['detection_masks'] =output_dict['detection_masks'][0] return output_dict
[9]:
import cv2cam =cv2.cv2.VideoCapture(0)rolling = Truewhile (rolling): ret,image_np = cam.read() image_np_expanded= np.expand_dims(image_np, axis=0) # Actual detection. output_dict= run_inference_for_single_image(image_np_expanded, detection_graph) # Visualization of the results of a detection. vis_util.visualize_boxes_and_labels_on_image_array( image_np, output_dict['detection_boxes'], output_dict['detection_classes'], output_dict['detection_scores'], category_index, instance_masks=output_dict.get('detection_masks'), use_normalized_coordinates=True, line_thickness=8) cv2.imshow('image', cv2.resize(image_np,(1000,800))) if cv2.waitKey(25) & 0xFF == ord('q'): break cv2.destroyAllWindows() cam.release()
在運行Jupyter notebook時，網絡攝影系統會開啟並檢測所有原始模型訓練過的物品類別。
感謝閱讀本文，如果有什麼建議，歡迎在留言區積極發言喲~
留言點讚關注
我們一起分享AI學習與發展的乾貨
如轉載，請後臺留言，遵守轉載規範