Object Detection utilites

Kano aims to support visualization and extract bounding boxes without the need to handle various box formats, such as xyxy, xywh, or scaled xywh:

• xywh2xyxy: convert box with xywh format (x_center, y_center, width, height - which is model input/output format) to xyxy format(x_min, y_min, x_max, y_max - which used to draw boxes).
• extract_bbox_area: get cropped box image from the image
• draw_bbox: draw bounding box on the image.

kano.detect_utils.xywh2xyxy(xywh)

Converts bounding box coordinates from (x_center, y_center, width, height) format to (x_min, y_min, x_max, y_max) format.

Parameters:

Name	Type	Description	Default
xywh	np.array) with shape (4,	A tuple containing (x_center, y_center, width, height) of the bounding box.	required

Returns:

Name	Type	Description
xyxy	tuple(int)	xyxy location of the bounding box.

Source code in kano\detect_utils.py

def xywh2xyxy(xywh):
    """
    Converts bounding box coordinates from (x_center, y_center, width, height) format to (x_min, y_min, x_max, y_max) format.

    Args:
        xywh (np.array) with shape (4,): A tuple containing (x_center, y_center, width, height) of the bounding box.

    Returns:
        xyxy (tuple(int)): xyxy location of the bounding box.
    """

    x_center, y_center, bbox_width, bbox_height = xywh
    x_min = int(x_center - bbox_width / 2)
    y_min = int(y_center - bbox_height / 2)
    x_max = int(x_center + bbox_width / 2)
    y_max = int(y_center + bbox_height / 2)

    return x_min, y_min, x_max, y_max

kano.detect_utils.extract_bbox_area(image, bbox)

Return cropped image from the given bounding box area

Parameters:

Name	Type	Description	Default
image	np.array) with shape (H, W, 3	image to extract the box	required
bbox	np.array) with shape (4,	xyxy location of the box	required

Returns:

Name	Type	Description
cropped_image	array	with shape (new_H, new_W, 3) based on bbox

Source code in kano\detect_utils.py

def extract_bbox_area(image, bbox):
    """
    Return cropped image from the given bounding box area

    Args:
        image (np.array) with shape (H, W, 3): image to extract the box
        bbox (np.array) with shape (4,): xyxy location of the box

    Returns:
        cropped_image (np.array): with shape (new_H, new_W, 3) based on bbox
    """

    (left, top), (right, bottom) = bbox[:2], bbox[2:]
    return image.copy()[top:bottom, left:right]

kano.detect_utils.draw_bbox(image, bbox, bbox_type='xyxy', bbox_color=(0, 0, 255), label=None)

Draws a bounding box on the image and optionally draws a multi-line label.

Parameters:

Name	Type	Description	Default
image	ndarray or str	The image on which the bounding box will be drawn.	required
bbox	list or tuple or ndarray	The bounding box coordinates. If it's a list, it should be in the format specified by bbox_type.	required
bbox_type	str	Type of bounding box coordinates. Should be either "xyxy" or "xywh" or "s_xywh".	'xyxy'
bbox_color	tuple	Color of the bounding box in BGR format.	(0, 0, 255)
label	str	Label to be displayed alongside the bounding box. Supports multiple lines with '/n' separating lines.	None

Returns:

Type	Description
ndarray	np.ndarray: Image with the bounding box and label drawn.

Source code in kano\detect_utils.py

def draw_bbox(
    image: Union[np.ndarray, str],
    bbox: Union[List[float], Tuple[float, ...], np.ndarray],
    bbox_type: str = "xyxy",
    bbox_color: Tuple[int, int, int] = (0, 0, 255),
    label: Optional[str] = None,
) -> np.ndarray:
    """
    Draws a bounding box on the image and optionally draws a multi-line label.

    Args:
        image (np.ndarray or str): The image on which the bounding box will be drawn.
        bbox (list or tuple or np.ndarray): The bounding box coordinates. If it's a list, it should be in the format specified by bbox_type.
        bbox_type (str): Type of bounding box coordinates. Should be either "xyxy" or "xywh" or "s_xywh".
        bbox_color (tuple): Color of the bounding box in BGR format.
        label (str, optional): Label to be displayed alongside the bounding box. Supports multiple lines with '/n' separating lines.

    Returns:
        np.ndarray: Image with the bounding box and label drawn.
    """
    if isinstance(image, str):
        temp_image = cv2.imread(image)
    else:
        temp_image = image.copy()

    image_height, image_width = temp_image.shape[:2]

    if isinstance(bbox, (list, tuple)):
        temp_bbox = np.array(bbox)
    temp_bbox = temp_bbox.copy()
    if "s_" in bbox_type:
        temp_bbox *= np.array(
            [image_width, image_height, image_width, image_height]
        )

    temp_bbox = temp_bbox.astype(np.int64)

    if "xyxy" in bbox_type:
        x_min, y_min, x_max, y_max = temp_bbox
    elif "xywh" in bbox_type:
        x_min, y_min, x_max, y_max = xywh2xyxy(temp_bbox)
    else:
        raise ValueError("Invalid bounding box type")

    cv2.rectangle(temp_image, (x_min, y_min), (x_max, y_max), bbox_color, 2)

    if label is not None:
        font = cv2.FONT_HERSHEY_SIMPLEX
        font_scale, thickness, pad = get_font_config(image_height)

        label_lines = label.split("\n")
        (text_width, text_height), _ = cv2.getTextSize(
            "sample", font, font_scale, thickness
        )
        y_offset = y_min - (text_height + pad) * (len(label_lines) - 1)

        for line in label_lines:
            (text_width, text_height), _ = cv2.getTextSize(
                line, font, font_scale, thickness
            )

            background_position = (x_min, y_offset)
            background_end_position = (
                x_min + text_width,
                y_offset - text_height - pad,
            )
            cv2.rectangle(
                temp_image,
                background_position,
                background_end_position,
                bbox_color,
                -1,
            )
            cv2.putText(
                temp_image,
                line,
                (x_min, y_offset - pad // 2),
                font,
                font_scale,
                (255, 255, 255),
                thickness,
            )

            y_offset += text_height + pad

    return temp_image