edgeboxes algorithm, api advice?

asked 2015-07-27 01:43:16 -0500

ultrafro gravatar image

Hi all,

I'd like to implement the edge-boxes algorithm for objectness (http://research.microsoft.com/pubs/22...), and I'm looking for some input on the api. I have a re-written version running on matlab (based on the paper and having looked at the author's matlab code).

1) Is anyone already working on this? 2) Where in OpenCV do you think this should go?

My thoughts are that this should be included into cv::saliency::Objectness. The algorithm first performs edge detection, then performs a search for windows in the image where contours are self-contained. Any input is very much welcomed!

edit retag flag offensive close merge delete



Actually I think this should go in the xobjectdetection module, because it is basically an approach, that based on image content (edge informations) tries to guess where some object classes can be found. This approach can then be used to do unsupervised learning on object classes if I am correct.

StevenPuttemans gravatar imageStevenPuttemans ( 2015-07-27 07:59:23 -0500 )edit

Thanks, I took a look at the objDetect module, however it still seems like saliency makes sense. ObjDetect module focuses on either classifying the object, or extracting features. EdgeBoxes, however, proposes whether a window is likely to have any object.

From Saliency class: "Objectness: Objectness is usually represented as a value which reflects how likely an image window covers an object of any category. Algorithms belonging to this category, avoid making decisions early on, by proposing a small number of category-independent proposals, that are expected to cover all objects in an image. Being able to perceive objects before identifying them is closely related to bottom up visual attention (saliency)."


ultrafro gravatar imageultrafro ( 2015-07-28 09:44:55 -0500 )edit

Any progress on the implementation? I'm very interested in it.

d11e gravatar imaged11e ( 2015-09-01 10:01:19 -0500 )edit

I've made some progress, but would really appreciate feedback:


ultrafro gravatar imageultrafro ( 2015-09-08 04:56:12 -0500 )edit

@ultrafro, I suggest you open up a PR with your current branch, so that an OpenCV core dev can jump in and help you where needed! Your progress seems very nice!

StevenPuttemans gravatar imageStevenPuttemans ( 2015-09-08 06:11:09 -0500 )edit

@ultrafro, I was interested in implementing the same research paper for my further use. Could you also help me with getting the orientation image from the edge image via NMS. Do you have the module for that?

shagun96 gravatar imageshagun96 ( 2017-06-21 05:19:03 -0500 )edit