从语音估计肢体动作

一般分几个流派：2d点派、3d点派以及3d旋转派，3d旋转派又分为欧拉角、四元数、rot6d三派，不确定哪个派别是效果是最好的？回头测试一下

def rotation_6d_to_matrix(d6: torch.Tensor) -> torch.Tensor:
    """
    Converts 6D rotation representation by Zhou et al. [1] to rotation matrix
    using Gram--Schmidt orthogonalization per Section B of [1].
    Args:
        d6: 6D rotation representation, of size (*, 6)
    Returns:
        batch of rotation matrices of size (*, 3, 3)
    [1] Zhou, Y., Barnes, C., Lu, J., Yang, J., & Li, H.
    On the Continuity of Rotation Representations in Neural Networks.
    IEEE Conference on Computer Vision and Pattern Recognition, 2019.
    Retrieved from http://arxiv.org/abs/1812.07035
    """
    a1, a2 = d6[..., :3], d6[..., 3:]
    b1 = F.normalize(a1, dim=-1)
    b2 = a2 - (b1 * a2).sum(-1, keepdim=True) * b1
    b2 = F.normalize(b2, dim=-1)
    b3 = torch.cross(b1, b2, dim=-1)
    return torch.stack((b1, b2, b3), dim=-2)

def matrix_to_rotation_6d(matrix: torch.Tensor) -> torch.Tensor:
    """
    Converts rotation matrices to 6D rotation representation by Zhou et al. [1]
    by dropping the last row. Note that 6D representation is not unique.
    Args:
        matrix: batch of rotation matrices of size (*, 3, 3)
    Returns:
        6D rotation representation, of size (*, 6)
    [1] Zhou, Y., Barnes, C., Lu, J., Yang, J., & Li, H.
    On the Continuity of Rotation Representations in Neural Networks.
    IEEE Conference on Computer Vision and Pattern Recognition, 2019.
    Retrieved from http://arxiv.org/abs/1812.07035
    """
    return matrix[..., :2, :].clone().reshape(*matrix.size()[:-2], 6)