0
$\begingroup$

In my project, I am detecting only one class, which is "airplane", using yolov5. However, at some frames, the neural network labels some of the buildings as airplanes, which obviously are not. This noise happens like 1 frame among 60 frames. How should I treat this issue? Which algorithms can be applied to filter out?

Improve this question
$\endgroup$
  • $\begingroup$ I had a similar issue with faces detection and Yolo. The best solution from an accuracy standpoint was in adding simple classification network for 2 classes 0-not a face and 1 -face. And passing resized block detected by you to double-check by this net. It perfectly filters almost all Yolo's false detections. Please let me know if you consider it can suit your case - I can share more details. $\endgroup$ – Stepan Novikov Nov 16 '20 at 10:43
  • $\begingroup$ Hello @StepanNovikov actually I think this is what I need. Best way to see that is to test and see I believe. I trained my object detection yolov5 with the repository https://github.com/ultralytics/yolov5. I might be wrong but what you means is adding some extra layers at the end of yolov5 network,right? $\endgroup$ – Berkant Ay Nov 16 '20 at 10:53
  • $\begingroup$ Can you please rephrase this sentence "However, at some frames network boxes, some of the building around my perspective which obviously not an airplane." so that it is clear? What does "frames network boxes" mean? Then you say "some of the building around my perspective", what your perspective? $\endgroup$ – nbro♦ Nov 16 '20 at 11:47
  • $\begingroup$ however at some of the frames, network tags some of the buildings as airplane which they are not. @nbro. Stepan understood btw. $\endgroup$ – Berkant Ay Nov 16 '20 at 12:05
  • 1
    $\begingroup$ Thanks @StepanNovikov I will try your solution and let you know. $\endgroup$ – Berkant Ay Nov 17 '20 at 11:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.