[ros-users] Detector Evaluation
edgar.riba at gmail.com
Wed Mar 2 10:19:04 UTC 2016
regarding detectors evaluation, one typical way to evaluate them in
research is using VLBenchmarks
<http://www.vlfeat.org/benchmarks/index.html>, a MATLAB framework for
testing image feature detectors and descriptors which some parts are
based on the original code
<http://www.robots.ox.ac.uk/%7Evgg/research/affine/#software> of K.
Mikolajczyk <http://www.inrialpes.fr/lear/people/Mikolajczyk>and others
just uses only Vlfeat. However, OpenCV has a slightly implementation in
C++ based on that mentioned original code [1
In my opinion, if you are using this OpenCV implementation then the code
you want to release should go in the existent vision_opencv
<http://wiki.ros.org/vision_opencv> which has some implemented OpenCV apps.
On 02/03/16 09:57, Maarten de Vries via ros-users wrote:
> On 19 February 2016 at 22:42, Peter Lorenz via ros-users
> <ros-users at lists.ros.org <mailto:ros-users at lists.ros.org>> wrote:
> I want to publish a package on the ros website.
> It is about evaluation detectors in Computer Vision.
> Usually, the detector spits out, if and where it found a desired
> You can read in XML files, which have saved the ground truth and
> says where the object really is located.
> The package pops up a window where you will see the rectangle of
> the ground truth and the rectangle of the detector. So you can
> see, if those rectangles are overlapping, the detector is right
> and saves the percentage in
> file. This list will be read in a script, which shows up a ROC
> curve of your database.
> Automatic evaluation of (changes to) vision algorithms would be very
> useful. I'm very interested to see your package.
> Please, help me to put this package on the ros webpage and I will
> write a tutorial for it.
> I have no experience publishing packages on the ROS website.
> However, the code is normally hosted elsewhere, like github. Would it
> be possible for you to publish the package publicly first (for example
> on github) and give us all the possibility to test it out? Undoubtedly
> that's a good first step towards getting it published on the ROS
> website too.
> Kind regards,
> ros-users mailing list
> ros-users at lists.ros.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the ros-users