0

I am using OpenCV's detect multi-scale. I do not understand however, whether the scale0 parameter refers to a downscale or an upscale factor. (Or both?). The documentation is here.

The documentation says that the scale0 parameter refers to the co-efficient of the window increase. Does this then mean that the image is only ever downsampled, for nLevels?

Thank you.

Spacey
  • 2,941
  • 10
  • 47
  • 63
  • From my experience I think you are right, the image is always downsampled (or the detection size is upsampled). I think thats because the trained detectors are typically quite small. – Micka Nov 23 '14 at 19:51
  • Thanks @Micka. It seems that they never *upsample* the *image* though, which ... I am not sure why not... – Spacey Nov 23 '14 at 21:31
  • I dont know, but as I said it might be that standard training is done on "minimal" detector size (dimensions like 64x64) so that upsampling the image might be very costly and inefficient. If you have an educated guess about an area where a smaller object might be you can still upscale the whole image or just that area manually. If you always want to upscale, just do it manually too – Micka Nov 24 '14 at 06:30

0 Answers0