latest/annotators/

[giscus[bot]]

annotators/

A set of easy-to-use utilities that will come in handy in any computer vision project.

https://supervision.roboflow.com/annotators/

[hafsaJChaudhry]

how can you use annotators for only certain bounding box id instances?

[SkalskiP]

Hi, @hafsaJChaudhry 👋🏻 In principle, it is simple - you must filter out Detections before annotating. But you need to know what you are looking for.

For example, this will only draw people:

import supervision as sv

image = ...
detections = sv.Detections(...)
detections = detections[detections.class_id == 0]

bounding_box_annotator = sv.BoundingBoxAnnotator()
annotated_frame = bounding_box_annotator.annotate(
    scene=image.copy(),
    detections=detections
)

For example, this will draw only an object with tracker_id = 11:

import supervision as sv

image = ...
detections = sv.Detections(...)
detections = detections[detections.tracker_id == 11]

bounding_box_annotator = sv.BoundingBoxAnnotator()
annotated_frame = bounding_box_annotator.annotate(
    scene=image.copy(),
    detections=detections
)

[casperdias]

After try example code of counting-crossing line object, i notice the object BBOX must all through the line so the counter is updated. can it be made that if only the bottom line of bbox passes through then the counter will increase??

[SkalskiP]

Hi @casperdias 👋🏻 Yup, it is possible! You need to adjust the value of triggering_anchors. Initiate your LineZone like this.

line_zone = sv.LineZone(
    start=<START_POINT_VALUE>,
    end=<END_POINT_VALUE>, 
    triggering_anchors=[sv.Position.BOTTOM_LEFT, sv.Position.BOTTOM_RIGHT]
)

This way, LineZone will only trigger if the bottom edge crosses the line. LineZone does not have the docs yet... but you can look at the code responsible for this behavior here.

[casperdias]

TYSM, I tried it and it worked. about the counter, is there any way to access the result of the LineZone.Trigger like true or false?

[SkalskiP]

You mean if you can access information if specific detections cross line in this specific video frame?

[casperdias]

yes I already found it so no problem

[SkalskiP]

That's awesome to hear!

[wcoolers]

Hi, is there a way I can customize the labels, e.g, put names of my choice on objects detected? Also, is there a way to implement object re-identification, in case an object goes out of frame and comes back in? Thanks a lot

[SkalskiP]

is there a way to implement object re-identification

It is on our long-term TODO list but, it is currently not possible with supervision.

is there a way I can customize the labels

Absolutely. All you need to do is pass the labels list as the argument to the LabelAnnotator.annotate method. The one important thing is that len(labels) must be equal to len(detections).

labels = [
    <YOUR_LABEL>
    for _
    in detections
]

label_annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER)
annotated_frame = label_annotator.annotate(
    scene=image.copy(),
    detections=detections,
    labels=labels
)

[casperdias]

Hi, is this library support protocol like RTSP or HTTP Live streaming to detect object?

[SkalskiP]

Supervision on its own does not yet support detection on live stream, but you can combine supervision with inference to do it.

[millzee01]

Hi, is there a way to copy the detected mask to a transparent image?

i.e create a transparent image and copy the mask object form the original image to the new transparent image thus making a new image contain just the masked detections?

[SkalskiP]

Hi @millzee01. Yes. Do you want to save each mask as separate image? Or put all together on the same image?

[SkalskiP]

Or do you prefer a result like this?

[tobieabel]

Hi - I've been using the HeatMap annotator and added in a simple homograph function so that xy coordinates from a typical CCTV camera view can be translated into a top down view for annotating onto a floor plan, as this is the most typical use-case I've found for heat mapping in industry. Works pretty well, and you can use your polygonZone webapp to generate the homography mappings. Is this something you would consider putting into supervision main branch?

[SkalskiP]

Hi @tobieabel 👋🏻 Would you be interested in preparing a PR with a POC of your solution?

[tobieabel]

@SkalskiP Sure, I'm a self taught amateur when it comes to coding and not done a PR before so you might need to guide me through it, but happy to submit the change I made, it only took a few lines of openCv code

[SkalskiP]

To start with, follow contributor guidelines. Using those instructions, you should be able to submit the first PR than I'll try to provide you with comments and further instructions.

[tobieabel]

@SkalskiP ok, I'll go through that and submit the PR this week and see what you think. Its just a small change and actually very similar to your transformation code on your blog on speed estimation on the 19th Jan this year

[SkalskiP]

@tobieabel awesome! I can't wait to take a look at the PR.

[aberg2323]

TraceAnnotator Question: how do i set the trace length as something other than the default. I am having trouble implementing this change in my code.

[LinasKo]

Gotcha. I'll verify tomorrow and get back to you.

[aberg2323]

Thank you so much for your help! I appreciate it greatly.

[LinasKo]

Hi @aberg2323,

I can't reproduce the issue. Here's the code I used - if there's an issue on your end, it may help debug it. And if the issue persists, let me know.

import supervision as sv
from supervision.assets import VideoAssets, download_assets
from ultralytics import YOLO


download_assets(VideoAssets.VEHICLES)

model = YOLO("yolov8s")

video_info = sv.VideoInfo.from_video_path(VideoAssets.VEHICLES.value)
frame_gen = sv.get_video_frames_generator(VideoAssets.VEHICLES.value)
tracker = sv.ByteTrack()
trace_annotator_1 = sv.TraceAnnotator(trace_length=45, color=sv.Color.BLUE)
trace_annotator_2 = sv.TraceAnnotator(trace_length=30, color=sv.Color.GREEN)
trace_annotator_3 = sv.TraceAnnotator(trace_length=15, color=sv.Color.WHITE)
trace_annotator_4 = sv.TraceAnnotator(trace_length=7, color=sv.Color.YELLOW)
trace_annotator_5 = sv.TraceAnnotator(trace_length=3, color=sv.Color.RED)


with sv.VideoSink("out.mp4", video_info) as sink:
    for frame in frame_gen:
        result = model(frame)[0]
        detections = sv.Detections.from_ultralytics(result)
        detections = tracker.update_with_detections(detections)
        annotated_frame = frame.copy()
        trace_annotator_1.annotate(annotated_frame, detections)
        trace_annotator_2.annotate(annotated_frame, detections)
        trace_annotator_3.annotate(annotated_frame, detections)
        trace_annotator_4.annotate(annotated_frame, detections)
        trace_annotator_5.annotate(annotated_frame, detections)
        sink.write_frame(annotated_frame)

[aberg2323]

Thank you so much! I was able to change the length with that code.

[LinasKo]

Best of luck in your project! 😉

[gusmendess]

I want to annotate my true positives as green bounding boxes, false positives as yellow bounding boxes and my false negatives as red bounding boxes for better understanding of my model. Of course comparing with my true bounding boxes annotations. How can i do that?

[SkalskiP]

Hi @gusmendess, do you mean something similar to this: https://x.com/skalskip92/status/1795194852121969104?

[gusmendess]

@SkalskiP Yes but in my case i want to compare my model predictions with my true labels. I want to my bounding boxes that are true positives to be color green, bboxes that are false positives color yellow and bboxes that are supposed to be detected but weren't as red bboxes