I’m building a computer vision pipeline where users upload multiple photos of the same physical scene taken from different angles (front view, side view, close-up) to build a clean list of unique objects and their attributes.
I’m running YOLO for object detection to extract bounding boxes and crop each detected object into lightweight ⁠.webp⁠ images. Because the same physical object appears across 3 or 4 different photos from different angles, I need to make sure each unique object is only counted once per session, and avoid running expensive downstream AI calls to re-describe an object we already processed from an earlier photo.