Function normalize_faces

imaginaire/model_utils/fs_vid2vid.py:565–628

r"""Normalize face keypoints w.r.t. the reference face keypoints. Args: keypoints (Kx2 numpy array): target facial keypoints. ref_keypoints (Kx2 numpy array): reference facial keypoints. Returns: keypoints (Kx2 numpy array): normalized facial keypoints.

(keypoints, ref_keypoints,
                    dist_scale_x=None, dist_scale_y=None)

def normalize_faces(keypoints, ref_keypoints,
                    dist_scale_x=None, dist_scale_y=None):
    r"""Normalize face keypoints w.r.t. the reference face keypoints.

    Args:
        keypoints (Kx2 numpy array): target facial keypoints.
        ref_keypoints (Kx2 numpy array): reference facial keypoints.
    Returns:
        keypoints (Kx2 numpy array): normalized facial keypoints.
    """
    if keypoints.shape[0] == 68:
        central_keypoints = [8]
        add_upper_face = False
        part_list = [[0, 16], [1, 15], [2, 14], [3, 13], [4, 12],
                     [5, 11], [6, 10], [7, 9, 8],
                     [17, 26], [18, 25], [19, 24], [20, 23], [21, 22],
                     [27], [28], [29], [30], [31, 35], [32, 34], [33],
                     [36, 45], [37, 44], [38, 43], [39, 42], [40, 47], [41, 46],
                     [48, 54], [49, 53], [50, 52], [51], [55, 59], [56, 58],
                     [57],
                     [60, 64], [61, 63], [62], [65, 67], [66]
                     ]
        if add_upper_face:
            part_list += [[68, 82], [69, 81], [70, 80], [71, 79], [72, 78],
                          [73, 77], [74, 76, 75]]
    elif keypoints.shape[0] == 126:
        central_keypoints = [16]
        part_list = [[i] for i in range(126)]
    else:
        raise ValueError('Input keypoints type not supported.')

    face_cen = np.mean(keypoints[central_keypoints, :], axis=0)
    ref_face_cen = np.mean(ref_keypoints[central_keypoints, :], axis=0)

    def get_mean_dists(pts, face_cen):
        r"""Get the mean xy distances of keypoints wrt face center."""
        mean_dists_x, mean_dists_y = [], []
        pts_cen = np.mean(pts, axis=0)
        for p, pt in enumerate(pts):
            mean_dists_x.append(np.linalg.norm(pt - pts_cen))
            mean_dists_y.append(np.linalg.norm(pts_cen - face_cen))
        mean_dist_x = sum(mean_dists_x) / len(mean_dists_x) + 1e-3
        mean_dist_y = sum(mean_dists_y) / len(mean_dists_y) + 1e-3
        return mean_dist_x, mean_dist_y

    if dist_scale_x is None:
        dist_scale_x, dist_scale_y = [None] * len(part_list), \
                                     [None] * len(part_list)

    for i, pts_idx in enumerate(part_list):
        pts = keypoints[pts_idx]
        if dist_scale_x[i] is None:
            ref_pts = ref_keypoints[pts_idx]
            mean_dist_x, mean_dist_y = get_mean_dists(pts, face_cen)
            ref_dist_x, ref_dist_y = get_mean_dists(ref_pts, ref_face_cen)

            dist_scale_x[i] = ref_dist_x / mean_dist_x
            dist_scale_y[i] = ref_dist_y / mean_dist_y