redesigned annotators

[SkalskiP]

This PR expands our range of Annotators and makes them more reusable.

[SkalskiP]

@hardikdava I created a feature branch that we can use to redesign annotators and make them more modular according to our discussion here #168

I decided that we will conduct all work around annotators on this feature branch and when we are done we will merge this branch as a whole into main.

[hardikdava]

Hi @SkalskiP 👋 , cool!. I will add modified box annotators in this branch.

[hardikdava]

@SkalskiP I have created initial PR as #171 . Let me know how can I contribute in this week.

[hardikdava]

@SkalskiP Related feature request to this PR #172

[hardikdava]

Test script to test new annotators.

import supervision as sv
import cv2
import numpy as np
from ultralytics import YOLO


img = cv2.imread("../data/zidane.jpg")

model = YOLO("yolov8s-seg.pt")

yolo_detect_result = model.predict([img], conf=0.5)[0]
detections = sv.Detections.from_yolov8(yolo_detect_result)

annotated_img = img.copy()

cat_maps = {0: "person", 27: "tie"}
labels = [f"{cat_maps[class_id]}" for _, _, _, class_id, _ in detections]

tracker_ids = np.array([1, 2, 3], dtype=int)
detections.tracker_id = tracker_ids  # Add tracker id to visualize color by tracker id


color_by_track = False   # Test with True and False
font_path = "font_path" # Test with actual path and None
skip_label = False  # Test with True and False

detection_annotator = sv.DetectionAnnotator(color_by_track=color_by_track, skip_label=skip_label)
segmentation_annotator = sv.SegmentationAnnotator(color_by_track=color_by_track, skip_label=skip_label)

box_annotator = sv.BoxAnnotator(color_by_track=color_by_track)
label_annotator = sv.LabelAnnotator(color_by_track=color_by_track)
mask_annotator = sv.MaskAnnotator(color_by_track=color_by_track)

boxmask_annotator = sv.BoxMaskAnnotator(color_by_track=color_by_track)
cornered_box_annotator = sv.CorneredBoxAnotator(color_by_track=color_by_track)
pil_label_annotator = sv.PillowLabelAnnotator(color_by_track=color_by_track, font=font_path )


annotators = {"label": label_annotator, "pill-label": pil_label_annotator, "box": box_annotator, "cornered_box": cornered_box_annotator, "box-mask": boxmask_annotator, "mask": mask_annotator, "detection":
              detection_annotator, "segmentation": segmentation_annotator}


for key, value in annotators.items():
    if key in ["label", "detection", "segmentation", "pill-label"]:
        annotated_img = annotators[key].annotate(scene=img.copy(), detections=detections, labels=labels)
    elif key == "pill-label":
        annotated_img = pil_label_annotator.annotate(scene=img.copy(), detections=detections, labels=labels)
    else:
        annotated_img = annotators[key].annotate(scene=img.copy(), detections=detections)
    cv2.imshow(key, annotated_img)
    cv2.waitKey(0)

[hardikdava]

@SkalskiP I made progress on this feature. PillowLabelAnnotator is only not implemented yet. All other annotators are implemented and tested locally.

[hardikdava]

Hi @SkalskiP 👋 I am planning to keep continue contribute to supervision. I have some questions 😕,

• How to update docs?
• Should we start using Github Projects? That will be useful for planning, prioritization, task assignment and keeping tracks on the issue.

[SkalskiP]

@hardikdava, how about meeting for 30 minutes on video call and talking through the priorities?

In principle, I would like us to be focused on those tasks in the first place: #181, but with all the loose ends I think meeting is the right move.

[hardikdava]

@SkalskiP Of course, we can talk. Let me know when we can have a short call.

[SkalskiP]

@hardikdava zoom or google meet?

[hardikdava]

@SkalskiP Google meet is fine.

[SkalskiP]

@hardikdava you are based in Germany? Can we meet at 8 PM? Or you'd like to do it earlier?

[hardikdava]

@SkalskiP Yeah, I am based in Germany. Would it possible to talk at 6 PM?

[SkalskiP]

@hardikdava, not today, but I could meet at 6 PM tomorrow. How about that?

[hardikdava]

@SkalskiP let's talk at 8 PM today then. I can make it.

[hardikdava]

@SkalskiP On side note, I am adding PillowLabelAnnotator(). Do we have any specific default font?

[SkalskiP]

@onuralpszr sure you can merge it in once I'm ready. I'm still not done, but tests are looking good. You can take a look here:

https://colab.research.google.com/drive/1iso4sYNNERe6KqH6zUixF7XwsapbCgzZ?usp=sharing

[SkalskiP]

@hardikdava and @onuralpszr take a look at the PR 🙏🏻

I dropped a few things from PR. We will add them to the next PRs. Still, we are getting most of the functionalities here.

I tested the changes here: https://colab.research.google.com/drive/1iso4sYNNERe6KqH6zUixF7XwsapbCgzZ?usp=sharing

[hardikdava]

Everything looks fine to me.

[onuralpszr]

I am still looking and adding something will post soon

[SkalskiP]

@onuralpszr let me know when you are ready ;)

[SkalskiP]

@onuralpszr, are we ready to merge?

[onuralpszr]

@hardikdava @SkalskiP thank you both for your awesome work. I am merging into develop let's continue make supervision even better.

[SkalskiP]

@onuralpszr I'm going to promote that feature hard this and next week!