Generally such points will have lower value, but this may not be accurate enough depending on your application. I think we got an accuracy close to 80% by playing with the threshold back in the days.
Detecting them beyound this would require re-training the model and adding for example an extra output to also predict wherever a point is obscured or not.
Generally such points will have lower value, but this may not be accurate enough depending on your application. I think we got an accuracy close to 80% by playing with the threshold back in the days. Detecting them beyound this would require re-training the model and adding for example an extra output to also predict wherever a point is obscured or not.