(Soft)NMS in Object Detection: PyTorch Implementation(with CUDA)

def soft_nms_cpu(
    np.ndarray[float, ndim=2] boxes_in,
    float iou_thr,
    unsigned int method=1,
    float sigma=0.5,
    float min_score=0.001,
):
    boxes = boxes_in.copy()
    cdef unsigned int N = boxes.shape[0]
    cdef float iw, ih, box_area
    cdef float ua
    cdef int pos = 0
    cdef float maxscore = 0
    cdef int maxpos = 0
    cdef float x1, x2, y1, y2, tx1, tx2, ty1, ty2, ts, area, weight, ov
    inds = np.arange(N)

    for i in range(N):
        maxscore = boxes[i, 4]
        maxpos = i

        tx1 = boxes[i, 0]
        ty1 = boxes[i, 1]
        tx2 = boxes[i, 2]
        ty2 = boxes[i, 3]
        ts = boxes[i, 4]
        ti = inds[i]

        pos = i + 1
        # get max box
        while pos < N:
            if maxscore < boxes[pos, 4]:
                maxscore = boxes[pos, 4]
                maxpos = pos
            pos = pos + 1

        # add max box as a detection
        boxes[i, 0] = boxes[maxpos, 0]
        boxes[i, 1] = boxes[maxpos, 1]
        boxes[i, 2] = boxes[maxpos, 2]
        boxes[i, 3] = boxes[maxpos, 3]
        boxes[i, 4] = boxes[maxpos, 4]
        inds[i] = inds[maxpos]

        # swap ith box with position of max box
        boxes[maxpos, 0] = tx1
        boxes[maxpos, 1] = ty1
        boxes[maxpos, 2] = tx2
        boxes[maxpos, 3] = ty2
        boxes[maxpos, 4] = ts
        inds[maxpos] = ti

        tx1 = boxes[i, 0]
        ty1 = boxes[i, 1]
        tx2 = boxes[i, 2]
        ty2 = boxes[i, 3]
        ts = boxes[i, 4]

        pos = i + 1
        # NMS iterations, note that N changes if detection boxes fall below
        # threshold
        while pos < N:
            x1 = boxes[pos, 0]
            y1 = boxes[pos, 1]
            x2 = boxes[pos, 2]
            y2 = boxes[pos, 3]
            s = boxes[pos, 4]

            area = (x2 - x1 + 1) * (y2 - y1 + 1)
            iw = (min(tx2, x2) - max(tx1, x1) + 1)
            if iw > 0:
                ih = (min(ty2, y2) - max(ty1, y1) + 1)
                if ih > 0:
                    ua = float((tx2 - tx1 + 1) * (ty2 - ty1 + 1) + area - iw * ih)
                    ov = iw * ih / ua  # iou between max box and detection box

                    if method == 1:  # linear
                        if ov > iou_thr:
                            weight = 1 - ov
                        else:
                            weight = 1
                    elif method == 2:  # gaussian
                        weight = np.exp(-(ov * ov) / sigma)
                    else:  # original NMS
                        if ov > iou_thr:
                            weight = 0
                        else:
                            weight = 1

                    boxes[pos, 4] = weight * boxes[pos, 4]

                    # if box score falls below threshold, discard the box by
                    # swapping with last box update N
                    if boxes[pos, 4] < min_score:
                        boxes[pos, 0] = boxes[N-1, 0]
                        boxes[pos, 1] = boxes[N-1, 1]
                        boxes[pos, 2] = boxes[N-1, 2]
                        boxes[pos, 3] = boxes[N-1, 3]
                        boxes[pos, 4] = boxes[N-1, 4]
                        inds[pos] = inds[N - 1]
                        N = N - 1
                        pos = pos - 1

            pos = pos + 1

    return boxes[:N], inds[:N]

def nms(dets, iou_thr, device_id=None):
    """Dispatch to either CPU or GPU NMS implementations.

    The input can be either a torch tensor or numpy array. GPU NMS will be used
    if the input is a gpu tensor or device_id is specified, otherwise CPU NMS
    will be used. The returned type will always be the same as inputs.

    Arguments:
        dets (torch.Tensor or np.ndarray): bboxes with scores.
        iou_thr (float): IoU threshold for NMS.
        device_id (int, optional): when `dets` is a numpy array, if `device_id`
            is None, then cpu nms is used, otherwise gpu_nms will be used.

    Returns:
        tuple: kept bboxes and indice, which is always the same data type as
            the input.
    """
    # convert dets (tensor or numpy array) to tensor
    if isinstance(dets, torch.Tensor):
        is_numpy = False
        dets_th = dets
    elif isinstance(dets, np.ndarray):
        is_numpy = True
        device = 'cpu' if device_id is None else 'cuda:{}'.format(device_id)
        dets_th = torch.from_numpy(dets).to(device)
    else:
        raise TypeError(
            'dets must be either a Tensor or numpy array, but got {}'.format(
                type(dets)))

    # execute cpu or cuda nms
    if dets_th.shape[0] == 0:
        inds = dets_th.new_zeros(0, dtype=torch.long)
    else:
        if dets_th.is_cuda:
            inds = nms_cuda.nms(dets_th, iou_thr)
        else:
            inds = nms_cpu.nms(dets_th, iou_thr)

    if is_numpy:
        inds = inds.cpu().numpy()
    return dets[inds, :], inds


def soft_nms(dets, iou_thr, method='linear', sigma=0.5, min_score=1e-3):
    if isinstance(dets, torch.Tensor):
        is_tensor = True
        dets_np = dets.detach().cpu().numpy()
    elif isinstance(dets, np.ndarray):
        is_tensor = False
        dets_np = dets
    else:
        raise TypeError(
            'dets must be either a Tensor or numpy array, but got {}'.format(
                type(dets)))

    method_codes = {'linear': 1, 'gaussian': 2}
    if method not in method_codes:
        raise ValueError('Invalid method for SoftNMS: {}'.format(method))
    new_dets, inds = soft_nms_cpu(
        dets_np,
        iou_thr,
        method=method_codes[method],
        sigma=sigma,
        min_score=min_score)

    if is_tensor:
        return dets.new_tensor(new_dets), dets.new_tensor(
            inds, dtype=torch.long)
    else:
        return new_dets.astype(np.float32), inds.astype(np.int64)