hogdescriptor::detectmultiscale to output feature vectors for detections (Feature #3848)

Added by Kevin Shi over 10 years ago. Updated over 9 years ago.

Status:Open Start date:2014-08-06
Priority:High Due date:
Assignee:- % Done:

0%
Category:objdetect
Target version:3.0
Difficulty: Pull request:

Description

detectMultiScale currently just outputs a vector<Rect> with no additional information. To get the feature vectors for each detection, I have to crop the image to each rect and compute the features again, which is extremely time consuming and slowing down my code by several factors.

Since these features are already computed in detectMultiScale, it should be possible to pass an additional argument (vector<vector<float> >& features) to detectMultiScale and have the detector write the feature vectors for each detection to this argument.

Associated revisions

Revision 7ab73229
Added by Vadim Pisarevsky almost 10 years ago

Merge pull request #3848 from yxiong:fix-4125

History

#1
Updated by Steven Puttemans over 10 years ago

I like the idea, but it should be implemented in the Algorithm class so that each descriptor detector can benefit from this.
You might be up for the job yourself and try it out? Making a pull request is quite straightforward and increase the chance your feature gets in.

#2
Updated by Dmitry Retinskiy over 10 years ago

  • Category set to objdetect
  • Target version changed from 3.0-beta to 3.0

#3
Updated by Deanna Hood almost 10 years ago

I'm interested in working on this. I am planning to overload each of the three existing detectMultiScale methods to have a version which provides the additional argument. Please advise if there is a preferred alternative to having 6 versions of the detectMultiScale function call.

#4
Updated by Steven Puttemans almost 10 years ago

Deanna Hood wrote:

I'm interested in working on this. I am planning to overload each of the three existing detectMultiScale methods to have a version which provides the additional argument. Please advise if there is a preferred alternative to having 6 versions of the detectMultiScale function call.

Since we are forced not to break backwards compatibility I think that overload functions with an extra parameter are the best way to go! If you are up to this in a pull request, then please also update the corresponding documentation so that people are aware of the new functionality.

  • Status changed from New to Open

#5
Updated by Deanna Hood almost 10 years ago

Just updating that I haven't gotten around to this, so that others know it's not assigned

#6
Updated by Maksim Shabunin over 9 years ago

Issue has been transferred to GitHub: https://github.com/Itseez/opencv/issues/4807

Also available in: Atom PDF