Is it possible to output multiple bounding boxes (distractors) similar to target

[shen338]

Thank you for the amazing work!

My task requires multiple outputs other than the target object. For example, if there are three cars that look similar in the search area, the tracker might have the target object with the highest score while the other two cars have a lower score.

Seems like the prediction head implementation only consider the result with the highest score: (https://github.com/researchmm/Stark/blob/565d255e4f8916d7a9977e0c1b02c4d4d67d5ec4/lib/models/stark/head.py#L56)

Is it possible to also output objects with a lower score? Will it work if I replace the prediction head of STARK with a detection prediction head?

Thank you!