Filter detections by position of bbox coordinates

[yieniggu]

Search before asking

• I have searched the Supervision issues and found no similar feature requests.

Question

Is there a way to exclude detections that are outside a given area of detection?

Something similar to:

detections = detections[detections.class_id == 0]

Currently I have a zone restricted as:

  top_start_point = (0, int(height * 0.2))
  top_end_point = (width, int(height * 0.2))
  bot_start_point = (0, int(height * 0.7))
  bot_end_point = (width, int(height * 0.7))

  frame = cv2.line(frame, top_start_point, top_end_point, (0, 255, 0), 4)
  frame = cv2.line(frame, bot_start_point, bot_end_point, (0, 255, 0), 4)

I tried something like

detections = detections[ [i for i in detections.xyxy if i[1] < height * 0.7 and i[1] > height * 0.2] ]

But it didn't work

Additional

No response

[github-actions[bot]]

Hello there, thank you for opening an Issue ! 🙏🏻 The team was notified and they will get back to you asap.

[SOBAN50]

@yieniggu There are different possibilities/options for that:

• Exclude detection if even only 1 pixel of detection is outside the restricted_zone i.e only include if whole detection box lies inside
• Include detection only if 50% (or a given percentage) or more of detection is inside the restricted_zone
• Include detection if even only a very small portion of detection lies inside

[SkalskiP]

Hi @yieniggu 👋🏻 Let me convert this question into a discussion and move it into the Q&A section.

[SkalskiP]

@yieniggu I think it is a perfect use-case for PolygonZone.

1. Define PolygonZone. I'm using the coordinates you defined above.

   import supervision as sv
   import numpy as np
   
   polygon = np.array([
       [0, height * 0.2],
       [width, height * 0.2)],
       [width, height * 0.7],
       [0, height * 0.7]
   ]).astype(int)
   
   zone = sv.PolygonZone(
       polygon=polygon,
       frame_resolution_wh=(width, height)
   )

2. Filter detections using zone as a condition. I'm going to assume you are using YOLOv8. Read more about from_ultralytics.

   import supervision as sv
   from ultralytics import YOLO
   
   model = YOLO('yolov8n.pt')
   result = model(image)[0]
   
   detections = sv.Detections.from_ultralytics(result)
   mask = zone.trigger(detections=detections)
   filtered_detections = detections[mask]

3. Read more about detection filtering here.

[yieniggu]

Thanks for the answer! Looks like it would fit perfectly. Couldn't find anything related to this kind of filter before. I'll give it a try and then come back with results

UPDATE: it worked great! Thanks again

[SkalskiP]

I'm so happy to hear that!

[yieniggu]

Just to heads up, the order of the vertexes in which you define a polygon (in case it's a rectangle or similar) should be top-left, top-right, bottom-right, bottom-left

:D