Function ResizedTensorShape

python/mod_cvcuda/operators/OpHQResize.cpp:216–270

};

inline Shape ResizedTensorShape(const nvcv::TensorLayout &srcLayout, const nvcv::TensorShape &srcShape,
                                const Shape &outShape)
{
    int resizeNDim = outShape.size();
    if (resizeNDim != 2 && resizeNDim != 3)
    {
        throw std::runtime_error(
            "The `out_shape` must be a tuple of 2 or 3 integers (for 2D or 3D resampling respectively).");
    }

    bool hasDepth     = srcLayout.find('D') >= 0;
    int  expectedNDim = hasDepth ? 3 : 2;

    if (expectedNDim != resizeNDim)
    {
        if (hasDepth)
        {
            throw std::runtime_error(
                "The input tensor contains depth extent (`D`) in the layout. For 3D resize, please specify the resized "
                "shape for 3 extents: depth, height, and width. Got 2 extents.");
        }
        else
        {
            throw std::runtime_error(
                "Expected the resized shape to consists of 2 integers: for resized height and width. Got 3 integers.");
        }
    }

    char shapeArgLayout[4] = "DHW";
    int  shapeArg[3];
    for (int d = 0; d < resizeNDim; d++)
    {
        shapeArg[d] = outShape[d].cast<int>();
    }

    Shape resizedShape(srcShape.rank());
    for (int i = 0; i < srcShape.rank(); i++)
    {
        resizedShape[i] = srcShape[i];
    }

    assert(srcShape.rank() == srcLayout.rank());
    for (int d = 0; d < resizeNDim; d++)
    {
        int axis = srcLayout.find(shapeArgLayout[d + 3 - resizeNDim]);
        if (axis < 0)
        {
            throw std::runtime_error(
                "The layout of an input tensor to the resize operator must contain HW extents in the layout (for "
                "images) or DHW extents (for 3D resampling). Some extents are missing in the input tensor.");
        }
        resizedShape[axis] = shapeArg[d];
    }
    return resizedShape;
}

class PyOpHQResize : public nvcvpy::Container
{