-1

I am trying to use opencv to train a cascade classifier. The object I am detecting has non-square shape. In training data set, the ROI I drew is non-squared (height > width). However, after training, the classifier outputs always have same width and height. Is there any option I can set to output non-square ROI? Thanks.

Jeru Luke
  • 20,118
  • 13
  • 80
  • 87
fnhdx
  • 321
  • 2
  • 5
  • 14

1 Answers1

2

choose width and height during training in the desired aspect ratio. OpenCV cascade classifiers can only detect the trained aspect ratio (and the chosen size is the minimum detectable size). Your samples are rescaled during training.

Micka
  • 19,585
  • 4
  • 56
  • 74