Vespa Velutina detection

edit

Hello,

I have some pictures taken in front of a beehive entry of Vespa Velutina, a bee predator.

(other images available at https://github.com/Jodaille/Jobee/blo...)

the annotations file: https://github.com/Jodaille/Jobee/blo...

I have made selections in this images :

opencv_annotation --annotations=annotations.txt --images=.

I have also negatives images: https://github.com/Jodaille/Jobee/tre...

and its file: https://github.com/Jodaille/Jobee/blo...

I have then created vectors file:

opencv_createsamples -info  annotations.txt -bg  negative.txt -vec positiveVectorFile.vec -w 10 -h 10

that give me : https://github.com/Jodaille/Jobee/blo...

and then using traincascade :

opencv_traincascade -data /home/jodaille/Jobee/OpenCv/data -vec positiveVectorFile.vec -bg negative.txt -numPos 10 -numNeg 10 -numStages 20  -w 10 -h 10

I have a OpenCv/data/cascade.xml result that I try to use with this script: https://github.com/Jodaille/Jobee/blo...

But it does not seem do detect the VV

I have tried different sizes changing : -w 10 -h 10 in createsamples and traincascade

Using opencv_visualisation with the casacade file it seems to analyse only a part of the image.

Are my images too big ? Should I use bigger width/height parameters ?