haosen9527/mobileNet-ssd

About object detection(mobileNet-ssd) tensorflow/language:c++

mobileNet-ssd

ABOUT

• mobilenet ssd
• tensorflow /c++/python
• catkin_make

INSTALL

 cd mobileNet-ssd
 catkin_make

RUN

• tensorflow c++

  rosrun mobileNet mobileNet_pb
  or
  ./devel/lib/mobileNet/mobileNet_pb

• opencv dnn

  rosrun mobileNet mobileNet_opencv
  or
  ./devel/lib/mobileNet/mobileNet_opencv

• python

  python src/mobileNet/python/ssd-mobilenet.py

tensorflow模型预测对比总结

将通过以下的方法进行模型预测部分的实现

• Tensorflow c++
• Tensorflow python
• Opencv dnn

对比分析内容

• 图片输入说明(图片加载)
  • Tensorflow c++ ：

    tensorflow ops::ReadFile/DecodePng

  • Tensorflow python ：

    load_image_into_numpy_array(numpy)

  • Opencv dnn :

    cv::imread(imagePath)

• 模型加载说明
  • Tensorflow c++ ：

    Status status = ReadBinaryProto(Env::Default(),MODEL_PATH,&graph_def);

  • Tensorflow python:

    with gfile.FastGFile(PATH_TO_PB) as f:
        graph_def = tf.GraphDef()
        graph_def.ParseFromString(f.read())
        sess.graph.as_default()
        tf.import_graph_def(graph_def,name='')
    or 
    with open(PATH_TO_PB) as f:
        graph_def = tf.GraphDef()
        .......

  • opencv DNN ：

    dnn::Net net = cv::dnn::readNetFromTensorflow(weights, prototxt);

时间对比：

• Opencvdnn ：detection time: 618.522 ms
• Tensorflow c++ ：detection time:699.195 ms
• Tensorflow python : detection time : 5916.458 ms

效果展示

• Tensorflow c++
  
• Tensorflow python
  
• Opencv dnn