Reference for ultralytics/utils/ops.py

def segment2box(segment, width: int = 640, height: int = 640):
    """
    Convert segment coordinates to bounding box coordinates.

    Converts a single segment label to a box label by finding the minimum and maximum x and y coordinates.
    Applies inside-image constraint and clips coordinates when necessary.

    Args:
        segment (torch.Tensor): Segment coordinates in format (N, 2) where N is number of points.
        width (int): Width of the image in pixels.
        height (int): Height of the image in pixels.

    Returns:
        (np.ndarray): Bounding box coordinates in xyxy format [x1, y1, x2, y2].
    """
    x, y = segment.T  # segment xy
    # Clip coordinates if 3 out of 4 sides are outside the image
    if np.array([x.min() < 0, y.min() < 0, x.max() > width, y.max() > height]).sum() >= 3:
        x = x.clip(0, width)
        y = y.clip(0, height)
    inside = (x >= 0) & (y >= 0) & (x <= width) & (y <= height)
    x = x[inside]
    y = y[inside]
    return (
        np.array([x.min(), y.min(), x.max(), y.max()], dtype=segment.dtype)
        if any(x)
        else np.zeros(4, dtype=segment.dtype)
    )  # xyxy

def scale_boxes(img1_shape, boxes, img0_shape, ratio_pad=None, padding: bool = True, xywh: bool = False):
    """
    Rescale bounding boxes from one image shape to another.

    Rescales bounding boxes from img1_shape to img0_shape, accounting for padding and aspect ratio changes.
    Supports both xyxy and xywh box formats.

    Args:
        img1_shape (tuple): Shape of the source image (height, width).
        boxes (torch.Tensor): Bounding boxes to rescale in format (N, 4).
        img0_shape (tuple): Shape of the target image (height, width).
        ratio_pad (tuple, optional): Tuple of (ratio, pad) for scaling. If None, calculated from image shapes.
        padding (bool): Whether boxes are based on YOLO-style augmented images with padding.
        xywh (bool): Whether box format is xywh (True) or xyxy (False).

    Returns:
        (torch.Tensor): Rescaled bounding boxes in the same format as input.
    """
    if ratio_pad is None:  # calculate from img0_shape
        gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1])  # gain  = old / new
        pad = (
            round((img1_shape[1] - img0_shape[1] * gain) / 2 - 0.1),
            round((img1_shape[0] - img0_shape[0] * gain) / 2 - 0.1),
        )  # wh padding
    else:
        gain = ratio_pad[0][0]
        pad = ratio_pad[1]

    if padding:
        boxes[..., 0] -= pad[0]  # x padding
        boxes[..., 1] -= pad[1]  # y padding
        if not xywh:
            boxes[..., 2] -= pad[0]  # x padding
            boxes[..., 3] -= pad[1]  # y padding
    boxes[..., :4] /= gain
    return clip_boxes(boxes, img0_shape)

def make_divisible(x: int, divisor):
    """
    Return the nearest number that is divisible by the given divisor.

    Args:
        x (int): The number to make divisible.
        divisor (int | torch.Tensor): The divisor.

    Returns:
        (int): The nearest number divisible by the divisor.
    """
    if isinstance(divisor, torch.Tensor):
        divisor = int(divisor.max())  # to int
    return math.ceil(x / divisor) * divisor

def nms_rotated(boxes, scores, threshold: float = 0.45, use_triu: bool = True):
    """
    Perform NMS on oriented bounding boxes using probiou and fast-nms.

    Args:
        boxes (torch.Tensor): Rotated bounding boxes with shape (N, 5) in xywhr format.
        scores (torch.Tensor): Confidence scores with shape (N,).
        threshold (float): IoU threshold for NMS.
        use_triu (bool): Whether to use torch.triu operator for upper triangular matrix operations.

    Returns:
        (torch.Tensor): Indices of boxes to keep after NMS.
    """
    sorted_idx = torch.argsort(scores, descending=True)
    boxes = boxes[sorted_idx]
    ious = batch_probiou(boxes, boxes)
    if use_triu:
        ious = ious.triu_(diagonal=1)
        # NOTE: handle the case when len(boxes) hence exportable by eliminating if-else condition
        pick = torch.nonzero((ious >= threshold).sum(0) <= 0).squeeze_(-1)
    else:
        n = boxes.shape[0]
        row_idx = torch.arange(n, device=boxes.device).view(-1, 1).expand(-1, n)
        col_idx = torch.arange(n, device=boxes.device).view(1, -1).expand(n, -1)
        upper_mask = row_idx < col_idx
        ious = ious * upper_mask
        # Zeroing these scores ensures the additional indices would not affect the final results
        scores[~((ious >= threshold).sum(0) <= 0)] = 0
        # NOTE: return indices with fixed length to avoid TFLite reshape error
        pick = torch.topk(scores, scores.shape[0]).indices
    return sorted_idx[pick]

def non_max_suppression(
    prediction,
    conf_thres: float = 0.25,
    iou_thres: float = 0.45,
    classes=None,
    agnostic: bool = False,
    multi_label: bool = False,
    labels=(),
    max_det: int = 300,
    nc: int = 0,  # number of classes (optional)
    max_time_img: float = 0.05,
    max_nms: int = 30000,
    max_wh: int = 7680,
    in_place: bool = True,
    rotated: bool = False,
    end2end: bool = False,
    return_idxs: bool = False,
):
    """
    Perform non-maximum suppression (NMS) on prediction results.

    Applies NMS to filter overlapping bounding boxes based on confidence and IoU thresholds. Supports multiple
    detection formats including standard boxes, rotated boxes, and masks.

    Args:
        prediction (torch.Tensor): Predictions with shape (batch_size, num_classes + 4 + num_masks, num_boxes)
            containing boxes, classes, and optional masks.
        conf_thres (float): Confidence threshold for filtering detections. Valid values are between 0.0 and 1.0.
        iou_thres (float): IoU threshold for NMS filtering. Valid values are between 0.0 and 1.0.
        classes (List[int], optional): List of class indices to consider. If None, all classes are considered.
        agnostic (bool): Whether to perform class-agnostic NMS.
        multi_label (bool): Whether each box can have multiple labels.
        labels (List[List[Union[int, float, torch.Tensor]]]): A priori labels for each image.
        max_det (int): Maximum number of detections to keep per image.
        nc (int): Number of classes. Indices after this are considered masks.
        max_time_img (float): Maximum time in seconds for processing one image.
        max_nms (int): Maximum number of boxes for torchvision.ops.nms().
        max_wh (int): Maximum box width and height in pixels.
        in_place (bool): Whether to modify the input prediction tensor in place.
        rotated (bool): Whether to handle Oriented Bounding Boxes (OBB).
        end2end (bool): Whether the model is end-to-end and doesn't require NMS.
        return_idxs (bool): Whether to return the indices of kept detections.

    Returns:
        output (List[torch.Tensor]): List of detections per image with shape (num_boxes, 6 + num_masks)
            containing (x1, y1, x2, y2, confidence, class, mask1, mask2, ...).
        keepi (List[torch.Tensor]): Indices of kept detections if return_idxs=True.
    """
    import torchvision  # scope for faster 'import ultralytics'

    # Checks
    assert 0 <= conf_thres <= 1, f"Invalid Confidence threshold {conf_thres}, valid values are between 0.0 and 1.0"
    assert 0 <= iou_thres <= 1, f"Invalid IoU {iou_thres}, valid values are between 0.0 and 1.0"
    if isinstance(prediction, (list, tuple)):  # YOLOv8 model in validation model, output = (inference_out, loss_out)
        prediction = prediction[0]  # select only inference output
    if classes is not None:
        classes = torch.tensor(classes, device=prediction.device)

    if prediction.shape[-1] == 6 or end2end:  # end-to-end model (BNC, i.e. 1,300,6)
        output = [pred[pred[:, 4] > conf_thres][:max_det] for pred in prediction]
        if classes is not None:
            output = [pred[(pred[:, 5:6] == classes).any(1)] for pred in output]
        return output

    bs = prediction.shape[0]  # batch size (BCN, i.e. 1,84,6300)
    nc = nc or (prediction.shape[1] - 4)  # number of classes
    nm = prediction.shape[1] - nc - 4  # number of masks
    mi = 4 + nc  # mask start index
    xc = prediction[:, 4:mi].amax(1) > conf_thres  # candidates
    xinds = torch.stack([torch.arange(len(i), device=prediction.device) for i in xc])[..., None]  # to track idxs

    # Settings
    # min_wh = 2  # (pixels) minimum box width and height
    time_limit = 2.0 + max_time_img * bs  # seconds to quit after
    multi_label &= nc > 1  # multiple labels per box (adds 0.5ms/img)

    prediction = prediction.transpose(-1, -2)  # shape(1,84,6300) to shape(1,6300,84)
    if not rotated:
        if in_place:
            prediction[..., :4] = xywh2xyxy(prediction[..., :4])  # xywh to xyxy
        else:
            prediction = torch.cat((xywh2xyxy(prediction[..., :4]), prediction[..., 4:]), dim=-1)  # xywh to xyxy

    t = time.time()
    output = [torch.zeros((0, 6 + nm), device=prediction.device)] * bs
    keepi = [torch.zeros((0, 1), device=prediction.device)] * bs  # to store the kept idxs
    for xi, (x, xk) in enumerate(zip(prediction, xinds)):  # image index, (preds, preds indices)
        # Apply constraints
        # x[((x[:, 2:4] < min_wh) | (x[:, 2:4] > max_wh)).any(1), 4] = 0  # width-height
        filt = xc[xi]  # confidence
        x, xk = x[filt], xk[filt]

        # Cat apriori labels if autolabelling
        if labels and len(labels[xi]) and not rotated:
            lb = labels[xi]
            v = torch.zeros((len(lb), nc + nm + 4), device=x.device)
            v[:, :4] = xywh2xyxy(lb[:, 1:5])  # box
            v[range(len(lb)), lb[:, 0].long() + 4] = 1.0  # cls
            x = torch.cat((x, v), 0)

        # If none remain process next image
        if not x.shape[0]:
            continue

        # Detections matrix nx6 (xyxy, conf, cls)
        box, cls, mask = x.split((4, nc, nm), 1)

        if multi_label:
            i, j = torch.where(cls > conf_thres)
            x = torch.cat((box[i], x[i, 4 + j, None], j[:, None].float(), mask[i]), 1)
            xk = xk[i]
        else:  # best class only
            conf, j = cls.max(1, keepdim=True)
            filt = conf.view(-1) > conf_thres
            x = torch.cat((box, conf, j.float(), mask), 1)[filt]
            xk = xk[filt]

        # Filter by class
        if classes is not None:
            filt = (x[:, 5:6] == classes).any(1)
            x, xk = x[filt], xk[filt]

        # Check shape
        n = x.shape[0]  # number of boxes
        if not n:  # no boxes
            continue
        if n > max_nms:  # excess boxes
            filt = x[:, 4].argsort(descending=True)[:max_nms]  # sort by confidence and remove excess boxes
            x, xk = x[filt], xk[filt]

        # Batched NMS
        c = x[:, 5:6] * (0 if agnostic else max_wh)  # classes
        scores = x[:, 4]  # scores
        if rotated:
            boxes = torch.cat((x[:, :2] + c, x[:, 2:4], x[:, -1:]), dim=-1)  # xywhr
            i = nms_rotated(boxes, scores, iou_thres)
        else:
            boxes = x[:, :4] + c  # boxes (offset by class)
            i = torchvision.ops.nms(boxes, scores, iou_thres)  # NMS
        i = i[:max_det]  # limit detections

        output[xi], keepi[xi] = x[i], xk[i].reshape(-1)
        if (time.time() - t) > time_limit:
            LOGGER.warning(f"NMS time limit {time_limit:.3f}s exceeded")
            break  # time limit exceeded

    return (output, keepi) if return_idxs else output

def clip_boxes(boxes, shape):
    """
    Clip bounding boxes to image boundaries.

    Args:
        boxes (torch.Tensor | numpy.ndarray): Bounding boxes to clip.
        shape (tuple): Image shape as (height, width).

    Returns:
        (torch.Tensor | numpy.ndarray): Clipped bounding boxes.
    """
    if isinstance(boxes, torch.Tensor):  # faster individually (WARNING: inplace .clamp_() Apple MPS bug)
        boxes[..., 0] = boxes[..., 0].clamp(0, shape[1])  # x1
        boxes[..., 1] = boxes[..., 1].clamp(0, shape[0])  # y1
        boxes[..., 2] = boxes[..., 2].clamp(0, shape[1])  # x2
        boxes[..., 3] = boxes[..., 3].clamp(0, shape[0])  # y2
    else:  # np.array (faster grouped)
        boxes[..., [0, 2]] = boxes[..., [0, 2]].clip(0, shape[1])  # x1, x2
        boxes[..., [1, 3]] = boxes[..., [1, 3]].clip(0, shape[0])  # y1, y2
    return boxes

def clip_coords(coords, shape):
    """
    Clip line coordinates to image boundaries.

    Args:
        coords (torch.Tensor | numpy.ndarray): Line coordinates to clip.
        shape (tuple): Image shape as (height, width).

    Returns:
        (torch.Tensor | numpy.ndarray): Clipped coordinates.
    """
    if isinstance(coords, torch.Tensor):  # faster individually (WARNING: inplace .clamp_() Apple MPS bug)
        coords[..., 0] = coords[..., 0].clamp(0, shape[1])  # x
        coords[..., 1] = coords[..., 1].clamp(0, shape[0])  # y
    else:  # np.array (faster grouped)
        coords[..., 0] = coords[..., 0].clip(0, shape[1])  # x
        coords[..., 1] = coords[..., 1].clip(0, shape[0])  # y
    return coords

def scale_image(masks, im0_shape, ratio_pad=None):
    """
    Rescale masks to original image size.

    Takes resized and padded masks and rescales them back to the original image dimensions, removing any padding
    that was applied during preprocessing.

    Args:
        masks (np.ndarray): Resized and padded masks with shape [H, W, N] or [H, W, 3].
        im0_shape (tuple): Original image shape as (height, width).
        ratio_pad (tuple, optional): Ratio and padding values as ((ratio_h, ratio_w), (pad_h, pad_w)).

    Returns:
        (np.ndarray): Rescaled masks with shape [H, W, N] matching original image dimensions.
    """
    # Rescale coordinates (xyxy) from im1_shape to im0_shape
    im1_shape = masks.shape
    if im1_shape[:2] == im0_shape[:2]:
        return masks
    if ratio_pad is None:  # calculate from im0_shape
        gain = min(im1_shape[0] / im0_shape[0], im1_shape[1] / im0_shape[1])  # gain  = old / new
        pad = (im1_shape[1] - im0_shape[1] * gain) / 2, (im1_shape[0] - im0_shape[0] * gain) / 2  # wh padding
    else:
        pad = ratio_pad[1]
    top, left = int(pad[1]), int(pad[0])  # y, x
    bottom, right = int(im1_shape[0] - pad[1]), int(im1_shape[1] - pad[0])

    if len(masks.shape) < 2:
        raise ValueError(f'"len of masks shape" should be 2 or 3, but got {len(masks.shape)}')
    masks = masks[top:bottom, left:right]
    masks = cv2.resize(masks, (im0_shape[1], im0_shape[0]))
    if len(masks.shape) == 2:
        masks = masks[:, :, None]

    return masks

def xyxy2xywh(x):
    """
    Convert bounding box coordinates from (x1, y1, x2, y2) format to (x, y, width, height) format where (x1, y1) is the
    top-left corner and (x2, y2) is the bottom-right corner.

    Args:
        x (np.ndarray | torch.Tensor): Input bounding box coordinates in (x1, y1, x2, y2) format.

    Returns:
        (np.ndarray | torch.Tensor): Bounding box coordinates in (x, y, width, height) format.
    """
    assert x.shape[-1] == 4, f"input shape last dimension expected 4 but input shape is {x.shape}"
    y = empty_like(x)  # faster than clone/copy
    y[..., 0] = (x[..., 0] + x[..., 2]) / 2  # x center
    y[..., 1] = (x[..., 1] + x[..., 3]) / 2  # y center
    y[..., 2] = x[..., 2] - x[..., 0]  # width
    y[..., 3] = x[..., 3] - x[..., 1]  # height
    return y

def xywh2xyxy(x):
    """
    Convert bounding box coordinates from (x, y, width, height) format to (x1, y1, x2, y2) format where (x1, y1) is the
    top-left corner and (x2, y2) is the bottom-right corner. Note: ops per 2 channels faster than per channel.

    Args:
        x (np.ndarray | torch.Tensor): Input bounding box coordinates in (x, y, width, height) format.

    Returns:
        (np.ndarray | torch.Tensor): Bounding box coordinates in (x1, y1, x2, y2) format.
    """
    assert x.shape[-1] == 4, f"input shape last dimension expected 4 but input shape is {x.shape}"
    y = empty_like(x)  # faster than clone/copy
    xy = x[..., :2]  # centers
    wh = x[..., 2:] / 2  # half width-height
    y[..., :2] = xy - wh  # top left xy
    y[..., 2:] = xy + wh  # bottom right xy
    return y

def xywhn2xyxy(x, w: int = 640, h: int = 640, padw: int = 0, padh: int = 0):
    """
    Convert normalized bounding box coordinates to pixel coordinates.

    Args:
        x (np.ndarray | torch.Tensor): Normalized bounding box coordinates in (x, y, w, h) format.
        w (int): Image width in pixels.
        h (int): Image height in pixels.
        padw (int): Padding width in pixels.
        padh (int): Padding height in pixels.

    Returns:
        y (np.ndarray | torch.Tensor): The coordinates of the bounding box in the format [x1, y1, x2, y2] where
            x1,y1 is the top-left corner, x2,y2 is the bottom-right corner of the bounding box.
    """
    assert x.shape[-1] == 4, f"input shape last dimension expected 4 but input shape is {x.shape}"
    y = empty_like(x)  # faster than clone/copy
    y[..., 0] = w * (x[..., 0] - x[..., 2] / 2) + padw  # top left x
    y[..., 1] = h * (x[..., 1] - x[..., 3] / 2) + padh  # top left y
    y[..., 2] = w * (x[..., 0] + x[..., 2] / 2) + padw  # bottom right x
    y[..., 3] = h * (x[..., 1] + x[..., 3] / 2) + padh  # bottom right y
    return y

def xyxy2xywhn(x, w: int = 640, h: int = 640, clip: bool = False, eps: float = 0.0):
    """
    Convert bounding box coordinates from (x1, y1, x2, y2) format to (x, y, width, height, normalized) format. x, y,
    width and height are normalized to image dimensions.

    Args:
        x (np.ndarray | torch.Tensor): Input bounding box coordinates in (x1, y1, x2, y2) format.
        w (int): Image width in pixels.
        h (int): Image height in pixels.
        clip (bool): Whether to clip boxes to image boundaries.
        eps (float): Minimum value for box width and height.

    Returns:
        (np.ndarray | torch.Tensor): Normalized bounding box coordinates in (x, y, width, height) format.
    """
    if clip:
        x = clip_boxes(x, (h - eps, w - eps))
    assert x.shape[-1] == 4, f"input shape last dimension expected 4 but input shape is {x.shape}"
    y = empty_like(x)  # faster than clone/copy
    y[..., 0] = ((x[..., 0] + x[..., 2]) / 2) / w  # x center
    y[..., 1] = ((x[..., 1] + x[..., 3]) / 2) / h  # y center
    y[..., 2] = (x[..., 2] - x[..., 0]) / w  # width
    y[..., 3] = (x[..., 3] - x[..., 1]) / h  # height
    return y

def xywh2ltwh(x):
    """
    Convert bounding box format from [x, y, w, h] to [x1, y1, w, h] where x1, y1 are top-left coordinates.

    Args:
        x (np.ndarray | torch.Tensor): Input bounding box coordinates in xywh format.

    Returns:
        (np.ndarray | torch.Tensor): Bounding box coordinates in xyltwh format.
    """
    y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
    y[..., 0] = x[..., 0] - x[..., 2] / 2  # top left x
    y[..., 1] = x[..., 1] - x[..., 3] / 2  # top left y
    return y

def xyxy2ltwh(x):
    """
    Convert bounding boxes from [x1, y1, x2, y2] to [x1, y1, w, h] format.

    Args:
        x (np.ndarray | torch.Tensor): Input bounding box coordinates in xyxy format.

    Returns:
        (np.ndarray | torch.Tensor): Bounding box coordinates in xyltwh format.
    """
    y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
    y[..., 2] = x[..., 2] - x[..., 0]  # width
    y[..., 3] = x[..., 3] - x[..., 1]  # height
    return y

def ltwh2xywh(x):
    """
    Convert bounding boxes from [x1, y1, w, h] to [x, y, w, h] where xy1=top-left, xy=center.

    Args:
        x (torch.Tensor): Input bounding box coordinates.

    Returns:
        (np.ndarray | torch.Tensor): Bounding box coordinates in xywh format.
    """
    y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
    y[..., 0] = x[..., 0] + x[..., 2] / 2  # center x
    y[..., 1] = x[..., 1] + x[..., 3] / 2  # center y
    return y

def xyxyxyxy2xywhr(x):
    """
    Convert batched Oriented Bounding Boxes (OBB) from [xy1, xy2, xy3, xy4] to [xywh, rotation] format.

    Args:
        x (numpy.ndarray | torch.Tensor): Input box corners with shape (N, 8) in [xy1, xy2, xy3, xy4] format.

    Returns:
        (numpy.ndarray | torch.Tensor): Converted data in [cx, cy, w, h, rotation] format with shape (N, 5).
            Rotation values are in radians from 0 to pi/2.
    """
    is_torch = isinstance(x, torch.Tensor)
    points = x.cpu().numpy() if is_torch else x
    points = points.reshape(len(x), -1, 2)
    rboxes = []
    for pts in points:
        # NOTE: Use cv2.minAreaRect to get accurate xywhr,
        # especially some objects are cut off by augmentations in dataloader.
        (cx, cy), (w, h), angle = cv2.minAreaRect(pts)
        rboxes.append([cx, cy, w, h, angle / 180 * np.pi])
    return torch.tensor(rboxes, device=x.device, dtype=x.dtype) if is_torch else np.asarray(rboxes)

def xywhr2xyxyxyxy(x):
    """
    Convert batched Oriented Bounding Boxes (OBB) from [xywh, rotation] to [xy1, xy2, xy3, xy4] format.

    Args:
        x (numpy.ndarray | torch.Tensor): Boxes in [cx, cy, w, h, rotation] format with shape (N, 5) or (B, N, 5).
            Rotation values should be in radians from 0 to pi/2.

    Returns:
        (numpy.ndarray | torch.Tensor): Converted corner points with shape (N, 4, 2) or (B, N, 4, 2).
    """
    cos, sin, cat, stack = (
        (torch.cos, torch.sin, torch.cat, torch.stack)
        if isinstance(x, torch.Tensor)
        else (np.cos, np.sin, np.concatenate, np.stack)
    )

    ctr = x[..., :2]
    w, h, angle = (x[..., i : i + 1] for i in range(2, 5))
    cos_value, sin_value = cos(angle), sin(angle)
    vec1 = [w / 2 * cos_value, w / 2 * sin_value]
    vec2 = [-h / 2 * sin_value, h / 2 * cos_value]
    vec1 = cat(vec1, -1)
    vec2 = cat(vec2, -1)
    pt1 = ctr + vec1 + vec2
    pt2 = ctr + vec1 - vec2
    pt3 = ctr - vec1 - vec2
    pt4 = ctr - vec1 + vec2
    return stack([pt1, pt2, pt3, pt4], -2)

def ltwh2xyxy(x):
    """
    Convert bounding box from [x1, y1, w, h] to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right.

    Args:
        x (np.ndarray | torch.Tensor): Input bounding box coordinates.

    Returns:
        (np.ndarray | torch.Tensor): Bounding box coordinates in xyxy format.
    """
    y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
    y[..., 2] = x[..., 2] + x[..., 0]  # width
    y[..., 3] = x[..., 3] + x[..., 1]  # height
    return y

def segments2boxes(segments):
    """
    Convert segment labels to box labels, i.e. (cls, xy1, xy2, ...) to (cls, xywh).

    Args:
        segments (list): List of segments where each segment is a list of points, each point is [x, y] coordinates.

    Returns:
        (np.ndarray): Bounding box coordinates in xywh format.
    """
    boxes = []
    for s in segments:
        x, y = s.T  # segment xy
        boxes.append([x.min(), y.min(), x.max(), y.max()])  # cls, xyxy
    return xyxy2xywh(np.array(boxes))  # cls, xywh

def resample_segments(segments, n: int = 1000):
    """
    Resample segments to n points each using linear interpolation.

    Args:
        segments (list): List of (N, 2) arrays where N is the number of points in each segment.
        n (int): Number of points to resample each segment to.

    Returns:
        (list): Resampled segments with n points each.
    """
    for i, s in enumerate(segments):
        if len(s) == n:
            continue
        s = np.concatenate((s, s[0:1, :]), axis=0)
        x = np.linspace(0, len(s) - 1, n - len(s) if len(s) < n else n)
        xp = np.arange(len(s))
        x = np.insert(x, np.searchsorted(x, xp), xp) if len(s) < n else x
        segments[i] = (
            np.concatenate([np.interp(x, xp, s[:, i]) for i in range(2)], dtype=np.float32).reshape(2, -1).T
        )  # segment xy
    return segments

def crop_mask(masks, boxes):
    """
    Crop masks to bounding box regions.

    Args:
        masks (torch.Tensor): Masks with shape (N, H, W).
        boxes (torch.Tensor): Bounding box coordinates with shape (N, 4) in relative point form.

    Returns:
        (torch.Tensor): Cropped masks.
    """
    _, h, w = masks.shape
    x1, y1, x2, y2 = torch.chunk(boxes[:, :, None], 4, 1)  # x1 shape(n,1,1)
    r = torch.arange(w, device=masks.device, dtype=x1.dtype)[None, None, :]  # rows shape(1,1,w)
    c = torch.arange(h, device=masks.device, dtype=x1.dtype)[None, :, None]  # cols shape(1,h,1)

    return masks * ((r >= x1) * (r < x2) * (c >= y1) * (c < y2))

def process_mask(protos, masks_in, bboxes, shape, upsample: bool = False):
    """
    Apply masks to bounding boxes using mask head output.

    Args:
        protos (torch.Tensor): Mask prototypes with shape (mask_dim, mask_h, mask_w).
        masks_in (torch.Tensor): Mask coefficients with shape (N, mask_dim) where N is number of masks after NMS.
        bboxes (torch.Tensor): Bounding boxes with shape (N, 4) where N is number of masks after NMS.
        shape (tuple): Input image size as (height, width).
        upsample (bool): Whether to upsample masks to original image size.

    Returns:
        (torch.Tensor): A binary mask tensor of shape [n, h, w], where n is the number of masks after NMS, and h and w
            are the height and width of the input image. The mask is applied to the bounding boxes.
    """
    c, mh, mw = protos.shape  # CHW
    ih, iw = shape
    masks = (masks_in @ protos.float().view(c, -1)).view(-1, mh, mw)  # CHW
    width_ratio = mw / iw
    height_ratio = mh / ih

    downsampled_bboxes = bboxes.clone()
    downsampled_bboxes[:, 0] *= width_ratio
    downsampled_bboxes[:, 2] *= width_ratio
    downsampled_bboxes[:, 3] *= height_ratio
    downsampled_bboxes[:, 1] *= height_ratio

    masks = crop_mask(masks, downsampled_bboxes)  # CHW
    if upsample:
        masks = F.interpolate(masks[None], shape, mode="bilinear", align_corners=False)[0]  # CHW
    return masks.gt_(0.0)