Function compute_valid_mask

script/utils/utils.py:305–331

Compute a boolean mask of the valid pixels resulting from an homography applied to an image of a given shape. Pixels that are False correspond to bordering artifacts. A margin can be discarded using erosion. Arguments: input_shape: Tensor of rank 2 representing the image sh

(image_shape, inv_homography, device='cpu', erosion_radius=0)

    return warped_img

def compute_valid_mask(image_shape, inv_homography, device='cpu', erosion_radius=0):
    """
    Compute a boolean mask of the valid pixels resulting from an homography applied to
    an image of a given shape. Pixels that are False correspond to bordering artifacts.
    A margin can be discarded using erosion.

    Arguments:
        input_shape: Tensor of rank 2 representing the image shape, i.e. `[H, W]`.
        homography: Tensor of shape (B, 8) or (8,), where B is the batch size.
        `erosion_radius: radius of the margin to be discarded.

    Returns: a Tensor of type `tf.int32` and shape (H, W).
    """

    if inv_homography.dim() == 2:
        inv_homography = inv_homography.view(-1, 3, 3)
    batch_size = inv_homography.shape[0]
    mask = torch.ones(batch_size, 1, image_shape[0], image_shape[1]).to(device)
    mask = inv_warp_image_batch(mask, inv_homography, device=device, mode='nearest')
    mask = mask.view(batch_size, image_shape[0], image_shape[1])
    mask = mask.cpu().numpy()
    if erosion_radius > 0:
        kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (erosion_radius*2,)*2)
        for i in range(batch_size):
            mask[i, :, :] = cv2.erode(mask[i, :, :], kernel, iterations=1)

    return torch.tensor(mask).to(device)

def Kalman1D(observations,damping=1):
    # To return the smoothed time series data