Is there a way to compute precision of a boosted cascade stages?

asked Mar 6 '18

Jad gravatar image

updated Aug 29 '18

So I have trained a boosted cascade using the Opencv utilities, I would like to evaluate the performance of each stage of the classifier. is there a way to access the results from individual stages or anything that could help achieving this?

Preview: (hide)

Comments

For now, the answer is quite straightforward, it is no. The internals make it quite hard to extract that information. I have been planning to do this for over a year and half now, but since then I kind of ran over to deep learning. What you could do, and what I have done, is make stage files manually (by deleting other stages), and then evaluatie those models 1 by 1. But it is quite cumbersome...

StevenPuttemans gravatar imageStevenPuttemans (Mar 6 '18)edit

Thank you Steven for the clarification I didnt think about that at all! I will probably do it as i only need it for once. I have also tried to visualize the cascade model to see what are the most discriminant Haar features in each stage but am having a problem so I have raised a question on: http://answers.opencv.org/question/18...

could you please help me answering it as well. Thank you in advance.

Jad gravatar imageJad (Mar 6 '18)edit