Advise needed for labeling overlapping objects

[Pino_Carenza]

Hello all,
in order to be consistent in my labeling jobs, I wanted to understand how to best label overlapping objects? Here is an example of my dilemma. I annotated the prey fully, but should I annotate the cat fully (even the non visible part of the mouth) or not?
Thank you for the advise

[Lenny]

Hello,

When it comes to labeling overlapping objects, the best practice is to label the entirety of each object, even if parts of it are occluded or not visible. This applies to your example as well; you should label the entire cat, including the non-visible part of the mouth. This approach helps your model understand the true bounds of objects and can improve its performance in detecting occluded objects in the future.

Here are some key points to remember while labeling:

1. Label entirely around the object. It’s better to include a small non-object buffer than to exclude a portion of the object with a rectangular label. This helps your model understand edges better (source).

2. For occluded objects, label them entirely. If an object is out of view due to another object being in front of it, label the object out of view as if you could see its entirety (source).

3. Beware of labeling many boxes that overlap or are entirely contained within each other. This can confuse your model (source).

4. If you’re worried about labeling occluded objects, you can label object detection models on Roboflow with polygons (and bounding boxes too). This can help localize the annotations/labels to the outline of the parts of the object that are visible (source).

Remember, the quality of your annotations is crucial for the success of your model. So, take your time and ensure you’re following these best practices. Happy labeling!

---

Hi! I’m Roboflow’s AI powered bot. This is an automated response powered using AI with knowledge from our docs, blog, and forum. If this solves your issue, please mark it as the solution using the checkbox icon. If this doesn’t solve your issue, please follow up with what you’ve tried/why, and the community will continue to chime in as usual.