Re: haartraining ...

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Re: haartraining ...

mshoaib_amin
Hi,
Anyone found the solution to this problem with performance.exe.

Regards

--- In [hidden email], "christian.kungel"
<Christian.Kungel@...> wrote:

>
> thanks, that's my command:
> - performance -data data/cascade -info testPositives.txt
> the testPositives.txt looks like this:
> image_1110.bmp
> image_1111.bmp
> image_1112.bmp
> image_1113.bmp ... and so on
>
> ok, you mean that i have to run the createsamples.exe in due to the
> testPositives? so i have to mark the objects with objectmarker.exe for
> this data set?
>
> until now, thanks a lot.
>
> christian
>
>
>
>
> --- In [hidden email], "michael_p_horton" <mhorton@> wrote:
> >
> > --- In [hidden email], "christian.kungel"
> > <Christian.Kungel@> wrote:
> > >
> > > hello everybody,
> > >
> > > at the moment, i work on my diploma thesis and i use OpenCV for
object
> > > detection.
> > >
> > > i want to detect cars from the back. so i've trained the opencv with
> > > such images of cars (haartraining). the images have not the same
size

> > > ... i hope this is ok at this point of view.
> > > the classifier is trained with 20 stages and the haartraining
> > > application returns with an ok.
> > > but now, if i run the performance application for detecting car
> > > objects from some sample images, i become an unexpected output ...
> > >
> > > here are the output sequence:
> > >
> > > +================================+======+======+======+
> > > |            File Name           | Hits |Missed| False|
> > > +================================+======+======+======+
> > > |                           Total|     0|     0|     0|
> > > +================================+======+======+======+
> >
> > It looks like you haven't run the performance application with a
> > proper list of files.   Remember, these have to be formatted to
> > specify positives in the same way as the input to createsamples.
> > Also, performance ignores any image with zero annotated positives in
> > it - if your sample images contain 0 annotations they'll be skipped.
> > You may wish to modify that in performance.cpp.
> >
> > When it's working properly, you should get something like this, with
> > one line per image file:
> > +================================+======+======+======+
> > |            File Name           | Hits |Missed| False|
> > +================================+======+======+======+
> > | SomeImageFilename.jpg          |    24|    60|     1|
> > +================================+======+======+======+
> > | SomeOtherImageFilename.jpg     |     3|     1|    42|
> > +================================+======+======+======+
> > | StillAnotherImageFilename.jpg  |     2|    99|     8|
> > +================================+======+======+======+
> > |                           Total|    29|   160|    51|
> > +================================+======+======+======+
> >
> > It should also create files called det-SomeImageFilename.jpg, etc.,
> > with boxes drawn around its detections.
> >
>