Unified Interface for Annotator

[hardikdava]

Currently, Annotators are divided individually BoxAnnotator and MaskAnnotator. It would be great to have a common interface by combining all the Annotators into one. This gives more simplified easy to use interface.

[SkalskiP]

Hi @hardikdava 👋🏻 Sorry, but we have them divided on purpose. We want annotators to have smaller responsibilities. I have experience with Annotators from previous projects I worked on. Keeping them separated is the smart move in long therm.

[hardikdava]

Hi @SkalskiP , I would suggest to keep individual and unified interface so the high-level usage will be use Unified Annotator and if a user wants to only use BoxAnnotator then they can use that alone. I know that having separate Annotator is much beneficial for example drawing mask when not needed will be waste of resource and time. But it is something nice to have.

[SkalskiP]

I'm not sure I understand. Could you write some pseudocode below?

[hardikdava]

@SkalskiP I tried to make a simple implementation with comments. Have a look here.

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

img = cv2.imread("bus.jog")
model = YOLO('yolov8s-seg.pt')

yolov8_results = model.predict(img)[0]

detections = sv.Detections.from_yolov8(yolov8_results)


# Current Implementation
# user has to deal with multiple annotators in usage
box_annotator = sv.BoxAnnotator()
mask_annotator = sv.MaskAnnotator()

annotated_img = box_annotator.annotate(scene=img, detections=detections)
annotated_img = mask_annotator.annotate(scene=annotated_img, detections=detections)

cv2.imshow("Result", annotated_img)
cv2.waitKey(0)


# Suggestion
# Part of Supervision
"""
Motivation: In future when supervision has capability to draw for multiple tasks,
eg. classfication, detection, segmentation, pose, 3d boxes, etc, then each application will have
individual annotator. 

Situation-1: when user wants to annotate every detection results-> boxes, track-id, masks, pose, etc
As a user: user has to deal with multiple annotators but instead only one interface

Situation-2: when user wants to annotate selected results -> only box and mask
User has to use needed annotators indvidiually or turn on/off using class attributes.


"""

class Annotator:

  def __init__(self):
    self.box_annotator = sv.BoxAnnotator()
    self.mask_annotator = sv.MaskAnnotator()

  def annotate(self, scene: np.ndarry, detections: sv.Detections)
    annotated_img = box_annotator.annotate(scene=img, detections=detections)
    if detections.mask:
        annotated_img = mask_annotator.annotate(scene=annotated_img, detections=detections)
    return annotated_img

# End of implementation


# as user application - only work with 2 lines in stead of multiple annotators
annotator = Annotator()
annotated_img = annotator(scene=img, detections=detections)

cv2.imshow("Result", annotated_img)
cv2.waitKey(0)

[hardikdava]

Closing this as #170 will have this solution.