Question

I have a bunch of different Feature classes which calculate image features.
I have to extract from these classes “key features” that will be used packed together as a search key.
And I will also store parts of the Feature classes. I can’t store entire feature classes because that would be quite inefficient.

Now, what I have thought of is writing a Stored_features class that puts together the “key features”.
My layout is:

   Facial_features
   |      |       |
   |      V       |
   | .---Feature1 V   <|-- Abstract_feature
   | | .---Feature2   <|---'
   V V V
  Stored_features

My problem is that such a Stored_features class would have a lot of getters and setters and as far as I know, getters and setters indicate bad design. Is there an easily maintainable way to avoid too many getters and setters here?

The point is I see my code get here very tightly coupled with this layout 🙁

EDIT:

My code extract as requested.

#include <opencv2/core/core.hpp>

class Abstract_feature{
public:
  virtual void calculate()=0;
  virtual void draw(cv::Mat& canvas)=0;
  /// to put values into Stored_features
  virtual void registrate_key_values(Stored_features&) const=0;
};

class Facial_features : Abstract_feature{
public:
  virtual void calculate()
  { 
    es.calculate; sc.calculate; 
    /*etc but iterating over a list of Abstract_feature's*/
  }
  Stored_features get_stored_features() const
  {
    return sf.clone();
  }
private:
  Stored_features sf;
  Head_size es;
  Skin_color sc;
};

class Head_size : public Abstract_feature{
  //you can guess the impl.
};    

class Stored_features{
public:
  typedef enum{SKIN_COLOR=0, HEAD_SIZE_WIDTH,HEAD_SIZE_HEIGHT} Name;
public:
  void set_key_feature(Name, double value);
  cv::Mat get_feature_vector() const {return key_values;}
private:
  cv::Mat key_values;
  // and here would come other features eg.
  cv::Rect head_roi; // I don't search based on these. (So I should not use getters/setters?)
};

Added opencv for it is an opencv-based project anyway.

Answer 1

The size of a class should be its own issue.

In this case I discovered that all I need is a pack() function for each feature class that removes all the data I do not want to store. Then I can store the classes as they are and don’t have to care about their sizes.