Labeling just one part of an object?

Hi all,

I have a question when it comes to labeling an object for Computer Vision / Object Detection.

I want to only detect from the torso and up over just the whole person since sometimes the person I can only see half their body and I really don’t care for legs.

What would be the best approach here?

Can I just label from the torso up, or does that make it harder for the object to be detected?

Everything I have watched states that they want the whole object while I only want half and there is not much out there about gathering half an object.


Hi @Meathelix, if your goal is to only detect the “torso and up,” rather than the whole person, you are fine to label just the torso and above.

One thing to be sure of: label all instances of people that appear in the images or sampled video frames from the “torso and up.” Any instance of these areas on a person that is visible in the image or video frame, even partially, should be labeled.

The key to identification of the object with higher precision is to ensure you are consistent with your labeling methodology.