Mastering OpenCV 3 - Second Edition

Training and visualization

An example program for training a shape model from the annotation data can be found in train_shape_model.cpp. With the command-line argument argv[1] containing the path to the annotation data, training begins by loading the data into memory and removing incomplete samples, as follows:

    ft_data data = load_ft<ft_data>(argv[1]); 
    data.rm_incomplete_samples();

The annotations for each example, and optionally their mirrored counterparts, are then stored in a vector before passing them to the training function as follows:

    vector<vector<Point2f>> points; 
    for(int i = 0; i < int(data.points.size()); i++) { 
      points.push_back(data.get_points(i,false)); 
      if(mirror)points.push_back(data.get_points(i,true)); 
    }

The shape model ...