search

chili-epfl / chilitags

Robust Fiducial Markers for Augmented Reality And Robotics
http://chili.epfl.ch/software
123 stars 57 forks source link

Fixing the mMaxInputWidth scaling bug #55

Closed lorenzolightsgdwarf closed 9 years ago

lorenzolightsgdwarf commented 10 years ago

Since the recognition doesn’t work with large images(~2000x1500), should we force the subsampling in that case?

qbonnard commented 10 years ago

Would you have an example of large image on which the recognition does not work ? I'm surprised that the resolution of the image has to do with it. In the test suite, we have bigger images: https://github.com/chili-epfl/chilitags-testdata/tree/master/stills/4360x2448

Or do you mean that the recognition does not work on images where tags are big ? But it would still work on large images with small tags ? That's an issue solved with the pyramidal subsampling, which is deactivated on the FAST performance preset. Then we could set the maxInputWidth as part of the FAST preset, because admittedly, processing 3Mpixels images in the search of 20x20 pixels is not exactly fast... But it would feel wrong to subsample by default too...

On Fri, Aug 29, 2014 at 10:26 AM, lorenzolightsndwarf < notifications@github.com> wrote:

Since the recognition doesn’t work with large images(~2000x1500), should we force the subsampling in that case?

— Reply to this email directly or view it on GitHub https://github.com/chili-epfl/chilitags/pull/55#issuecomment-53849826.

lorenzolightsgdwarf commented 10 years ago

Yes you are right: big tags + big image without pyramid.... displaychilitags_screenshot_29 08 2014