I am working on this project in which I have to detect objects only when they are complete(not when objects are incomplete or partially visible).I have more than 20 classes of objects.
I tried with yolo by annotating only complete objects, but that didn't work.

How can I approach this problem, any suggestion is useful.

  • 1
    $\begingroup$ This is an interesting question, as the usual goal is to generalise to include incomplete images. I don't have very much experience with YOLO, so not writing a full answer. But you could try this: Treating complete and incomplete objects of same type as different classes. This would allow you to continue using YOLO architecture, although I don't know how well it will cope with differentiating between such similar classes in practice - it should be step up from removing class labels altogether from training data. $\endgroup$ Commented Oct 11, 2019 at 13:04


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.