Function _prepare_input_pose

detrsmpl/core/visualization/visualize_smpl.py:204–249

Prepare input pose data as tensor and ensure correct temporal slice.

(verts, poses, betas, transl)

def _prepare_input_pose(verts, poses, betas, transl):
    """Prepare input pose data as tensor and ensure correct temporal slice."""
    if verts is None and poses is None:
        raise ValueError('Please input valid poses or verts.')
    elif (verts is not None) and (poses is not None):
        warnings.warn('Redundant input, will take verts and ignore poses & '
                      'betas & transl.')
        poses = None
        transl = None
        betas = None
    elif isinstance(poses, dict):
        transl = poses.get('transl', transl)
        betas = poses.get('betas', betas)

    if isinstance(verts, np.ndarray):
        verts = torch.Tensor(verts)
        num_frames = verts.shape[0]
    elif isinstance(verts, torch.Tensor):
        num_frames = verts.shape[0]

    if isinstance(poses, np.ndarray):
        poses = torch.Tensor(poses)
        num_frames = poses.shape[0]
    elif isinstance(poses, torch.Tensor):
        num_frames = poses.shape[0]
    elif isinstance(poses, dict):
        for k, v in poses.items():
            if isinstance(v, np.ndarray):
                poses[k] = torch.tensor(v)
        num_frames = poses['body_pose'].shape[0]

    if isinstance(betas, np.ndarray):
        betas = torch.Tensor(betas)

    if betas is not None:
        if betas.shape[0] != num_frames:
            times = num_frames // betas.shape[0]
            if betas.ndim == 2:
                betas = betas.repeat(times, 1)[:num_frames]
            elif betas.ndim == 3:
                betas = betas.repeat(times, 1, 1)[:num_frames]
            print(f'betas will be repeated by dim 0 for {times} times.')
    if isinstance(transl, np.ndarray):
        transl = torch.Tensor(transl)

    return verts, poses, betas, transl


def _prepare_mesh(poses, betas, transl, verts, start, end, body_model):