Simple tool using OpenCV to manually draw bounding boxes on an image, and return the coordinates of those boxes

draw_bounding_boxes.py

from typing import Optional, Tuple, Sequence
import cv2


def draw_bboxes_on_image(
        image,  # A (height, width, 3) unit8 numpy array representing an image
        return_relative=False,  # True to return box coorinates relative to width/height of image so they will be in [0, 1] interval
        window_name='Draw Your Bounding Box',  # What to call the window
        display_color=(0, 0, 255),  # Colour of the displayed box
        display_thickness=2  # Thickness of the displayed box
) -> Sequence[Tuple[float, float, float, float]]:
    """
    Draw a bounding box on the image, returning a (x_left, x_right, y_top, y_bottom) coordinates.

    Once this function is launched, an image will pop up.
    - Click and drag to add a new bounding box
    - Click 'Return' to close the window and return the currently shown boxes
    - Click 'c' (for clear) to clear all boxes and start again
    - Click 'b' (for back) to remove the last drawn box
    - Click 'q' (for quit) to close the window and return no boxes
    """

    class CallbackContainer:

        bboxes = []
        start_xy: Optional[Tuple[float, float]] = None

        @classmethod
        def on_click(cls, event, x, y, flags, param):
            print(f'Got event type {cv2.EVENT_LBUTTONDOWN} at {x}, {y}')
            if event == cv2.EVENT_LBUTTONDOWN:
                if cls.start_xy is None:
                    cls.start_xy = (x, y)
            elif event == cv2.EVENT_LBUTTONUP:
                xs, xe = sorted([cls.start_xy[0], x])
                ys, ye = sorted([cls.start_xy[1], y])
                cls.bboxes.append([xs, xe, ys, ye])

                print(f'Added box #{len(cls.bboxes)}: {cls.bboxes[-1]}')
                cls.show_display_image()
                cls.start_xy = None

        @classmethod
        def show_display_image(cls):
            display_image = image.copy()
            for xs, xe, ys, ye in cls.bboxes:
                cv2.rectangle(display_image, (xs, ys), (xe, ye), color=display_color, thickness=display_thickness)
            cv2.imshow(window_name, display_image)

    cv2.namedWindow(window_name)
    cv2.setMouseCallback(window_name, CallbackContainer.on_click)
    CallbackContainer.show_display_image()

    while True:
        print('Click and drag to make a box on the image')
        key = cv2.waitKey()
        if key == ord('c'):
            print('Clearted all boxes')
            CallbackContainer.bboxes.clear()
        elif key == ord('b'):
            print('Clearing last box')
            if len(CallbackContainer.bboxes)>0:
                CallbackContainer.bboxes.pop(-1)
        elif key == ord('q'):
            print('Quitting, returning no boxes')
            CallbackContainer.bboxes.clear()
            break
        elif key == 13:  # Enter
            print(f'Done, returning {len(CallbackContainer.bboxes)} boxes.')
            break
        else:
            print(f'Unknown keypress: {key}')

        CallbackContainer.show_display_image()

    cv2.destroyWindow(window_name)
    boxes = CallbackContainer.bboxes.copy()
    if return_relative:
        h, w = image.shape[:2]
        boxes = [[xs / w, xe / w, ys / h, ye / h] for xs, xe, ys, ye in boxes]
    print(f'Your boxes are\n  {boxes}')
    return boxes


if __name__ == "__main__":
    draw_bboxes_on_image(
        image=cv2.imread(cv2.samples.findFile('starry_night.jpg'))
    )