Function resize

monai/transforms/spatial/functional.py:311–382

Functional implementation of resize. This function operates eagerly or lazily according to ``lazy`` (default ``False``). Args: img: data to be changed, assuming `img` is channel-first. out_size: expected shape of spatial dimensions after resize operation. mo

(
    img, out_size, mode, align_corners, dtype, input_ndim, anti_aliasing, anti_aliasing_sigma, lazy, transform_info
)

def resize(
    img, out_size, mode, align_corners, dtype, input_ndim, anti_aliasing, anti_aliasing_sigma, lazy, transform_info
):
    """
    Functional implementation of resize.
    This function operates eagerly or lazily according to
    ``lazy`` (default ``False``).

    Args:
        img: data to be changed, assuming `img` is channel-first.
        out_size: expected shape of spatial dimensions after resize operation.
        mode: {``"nearest"``, ``"nearest-exact"``, ``"linear"``,
            ``"bilinear"``, ``"bicubic"``, ``"trilinear"``, ``"area"``}
            The interpolation mode.
            See also: https://pytorch.org/docs/stable/generated/torch.nn.functional.interpolate.html
        align_corners: This only has an effect when mode is
            'linear', 'bilinear', 'bicubic' or 'trilinear'.
        dtype: data type for resampling computation. If None, use the data type of input data.
        input_ndim: number of spatial dimensions.
        anti_aliasing: whether to apply a Gaussian filter to smooth the image prior
            to downsampling. It is crucial to filter when downsampling
            the image to avoid aliasing artifacts. See also ``skimage.transform.resize``
        anti_aliasing_sigma: {float, tuple of floats}, optional
            Standard deviation for Gaussian filtering used when anti-aliasing.
        lazy: a flag that indicates whether the operation should be performed lazily or not
        transform_info: a dictionary with the relevant information pertaining to an applied transform.
    """
    img = convert_to_tensor(img, track_meta=get_track_meta())
    orig_size = img.peek_pending_shape() if isinstance(img, MetaTensor) else img.shape[1:]
    extra_info = {
        "mode": mode,
        "align_corners": align_corners if align_corners is not None else TraceKeys.NONE,
        "dtype": str(dtype)[6:],  # dtype as string; remove "torch": torch.float32 -> float32
        "new_dim": len(orig_size) - input_ndim,
    }
    meta_info = TraceableTransform.track_transform_meta(
        img,
        sp_size=out_size,
        affine=scale_affine(orig_size, out_size, align_corners=align_corners if align_corners is not None else False),
        extra_info=extra_info,
        orig_size=orig_size,
        transform_info=transform_info,
        lazy=lazy,
    )
    if lazy:
        if anti_aliasing and lazy:
            warnings.warn("anti-aliasing is not compatible with lazy evaluation.")
        out = _maybe_new_metatensor(img)
        return out.copy_meta_from(meta_info) if isinstance(out, MetaTensor) else meta_info
    if tuple(convert_to_numpy(orig_size)) == out_size:
        out = _maybe_new_metatensor(img, dtype=torch.float32)
        return out.copy_meta_from(meta_info) if isinstance(out, MetaTensor) else out
    out = _maybe_new_metatensor(img)
    img_ = convert_to_tensor(out, dtype=dtype, track_meta=False)  # convert to a regular tensor
    if anti_aliasing and any(x < y for x, y in zip(out_size, img_.shape[1:])):
        factors = torch.div(torch.Tensor(list(img_.shape[1:])), torch.Tensor(out_size))
        if anti_aliasing_sigma is None:
            # if sigma is not given, use the default sigma in skimage.transform.resize