latest/how_to/detect_and_annotate/

[giscus[bot]]

latest/how_to/detect_and_annotate/

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

https://supervision.roboflow.com/latest/how_to/detect_and_annotate/

[artinm89]

this is so good but the annotating spend lot of space

[artinm89]

if there is any way for this issue pls let me know

[SkalskiP]

Hi @artinm89 👋🏻 Did you faced any specific speed related issue?

[AnatolySamaris]

how do I annotate gray image as it has less dimension?

[SkalskiP]

Unfortunately, we do not support grayscale image annotations.

[Atef98]

How to annotate multiple labels like class_name, tracker_id, confidence, ....?

[SkalskiP]

Hi @atef92! 👋🏻 Sorry for late response. Take a look here.

labels = [
    f"#{tracker_id} {model.model.names[class_id]} {confidence:.2f}"
    for class_id, confidence, tracker_id
    in zip(detections.class_id, detections.confidence, detections.tracker_id)
]

annotated_image = label_annotator.annotate(
    scene=annotated_image, detections=detections, labels=labels)

[MubashirWaheed]

For some reason only bounding box annotation is working for me

from ultralytics import YOLO
import cv2 as cv 
import supervision as sv
coco_model = YOLO("yolov8n.pt")

image = cv.imread("./cars.png")
result = coco_model(image)[0]
# parse detections
detections = sv.Detections.from_ultralytics(result)
print('detections: ',detections)

detections = detections[detections.class_id == 2]
bounding_box_annotator = sv.MaskAnnotator()
label_annotator = sv.LabelAnnotator()

labels = [
    coco_model.model.names[class_id]
    for class_id
    in detections.class_id
]
annotated_image = bounding_box_annotator.annotate(scene=image, detections=detections)
annotated_image = label_annotator.annotate(scene=annotated_image, detections=detections, labels=labels)

sv.plot_image(annotated_image)

what is wrong in this code?

[LinasKo]

Hi @MubashirWaheed 👋

detections you get from from_ultralytics is a universal object that can hold both object detections results (boxes) and segmentation results (masks). Some annotators only work on segmentation models.

Try changing to yolov8n-seg.pt and see what happens 🙂

[luispatriciorf]

Hi guys,
How can I modify the label colors based on classes? I trained a model using yolov8 but I realized there were two classes that were the same like "player" and "players". When I use the annotation feature it assigns different colors which makes sense (they're two different classes) but both classes would have to have the same color.

I didn't find anything about that.

Thanks in advance!

Luis.

[SkalskiP]

To clarify Option 1 further: Every annotator in Supervision accepts a color argument. This can be a single sv.Color, in which case all detections will be marked identically, or an sv.ColorPalette — a list of colors where each class is painted a different color. Referring to the example provided by Linus, if "player" and "players" are classes 1 and 8, you could define a custom sv.ColorPalette where the colors at indices 1 and 8 are the same and then pass it to your annotator. Here's an example:

import supervision as sv

color_palette = sv.ColorPalette.from_hex([
    "A351FB",
    "FF4040", # <--- color with index 1
    "FFA1A0",
    "FF7633",
    "FFB633",
    "D1D435",
    "4CFB12",
    "94CF1A",
    "FF4040", # <--- color with index 8
    "1B9640",
])

bounding_box_annotator = sv.BoundingBoxAnnotator(color=color_palette)

[luispatriciorf]

Awesome! Thanks guys!

[luispatriciorf]

Guys,
It's me again, sorry. I don't know if it is a dumb question but how can I modify the ellipse diameter if I use one? Not the thickness but the diameter.

sv.EllipseAnnotator() only shows me thickness, start and end angle, and color.

Thanks!

[LinasKo]

It's alright - you have a problem you want to solve, so there's no dumb questions :)

Edit: use SkalskiP's method.

---

Indeed, we don't have a scale factor. The elipse size depends on the detection size.
To save time, here's a very scrappy workaround - it will depend on your use case, whether it's acceptable:

You can modify the xyxy values of the detections, right before annotation:

extra_diameter = 5
detections.xyxy[:, 0] -= extra_diameter
detections.xyxy[:, 1] -= extra_diameter
detections.xyxy[:, 2] += extra_diameter
detections.xyxy[:, 3] += extra_diameter

This should add additional pixels on each side. But again - this is very hacky and won't work for all use cases.
Does it solve yours?

[SkalskiP]

Or, you can use pad_boxes util.

detections.xyxy = sv.pad_boxes(xyxy=detections.xyxy, px=5)

[pramudithaSJ]

This code work fine in my laptop , but in my raspberry pi 4 gives below error

labels = [ ^ TypeError: 'NoneType' object is not iterable

    while True:
        success, frame = self.cap.read()
        if success:
            results = self.yolo(frame)[0]
            detections = sv.Detections.from_ultralytics(results)
            detections = detections[detections.confidence > 0.70]

            if len(detections) > 0:
                print(detections)
                labels = [
                    class_name.split()[0]  # Extract just the numeric part
                    for class_name in detections['class_name']
                ]
                self.release_resources()
                return jsonify({"message": "High confidence detections found","data":labels}), 200

[LinasKo]

Hi @pramudithaSJ 👋

Checking the code, detections['class_name'] can return None. However, we do expect to see a names array in Ultralytics and create our own - it's strange. Could you please:

1. I assume this fails in class_name.split()[0] for class_name in detections['class_name']. Is that correct?
2. Verify you're using the same code, including the model number, the runtime (e.g. ultralytics, rather than inference)
3. Can you show me what's inside the detections object that fails? print(detections) should work for that.
4. Which model are you using? Ideally, seeing more code would be great.

[SkalskiP]

@pramudithaSJ, what version of supervision and ultralytics have you installed on your laptop and Raspberry Pi 4? What version of Python you have on both of those devices?

echo "Python version:"
python3 --version

# Check the supervision version
echo "supervision version:"
pip show supervision | grep Version

# Check the ultralytics version
echo "ultralytics version:"
pip show ultralytics | grep Version