After forward, do something like output bvh, get error value

copy an arbitrary bvh file as a static information (skeleton's offset) reference

Parameters ---------- parents : (J) ndarray parents array Returns ------- descendants : [nda

Animation structure edges Parameters ---------- parents : (J) ndarray parents array Returns

Extract skinning weight from a given mesh

If root, identity, or already generated, return

positions: [T, J, 3], trimmed (only "chosen_joints") fid_l, fid_r: indices of feet joints (in "chosen_joints")

if not quater and rotation.shape[-2] != 3: raise Exception('Unexpected shape of rotation') if quater and rotation.shape[-2] != 4: rai

Run forward pass; called by both functions <optimize_parameters> and <test>.

data: {"train": dict_train, "test": dict_test} dict_train: {"style2d_code": blabla, etc.}

Get the unit vectors for local rectangular coordinates for given 3D motion :param local_x: local x axis, (B *) [*, 0, *] :return: numpy a

input: positions - glb [T, J, (3/2)] -- single clip! output: motion with average root (x(, z)) = (0(, 0))

Find parent rotations

motion: motion in relative joint positions & global root positions [(B,) T, (J - 1) + 1, 3] local_x: [(B,) 3], local x-axi

Calculate losses, gradients, and update network weights; called in every training iteration

Parameters ---------- parents : (J) ndarray parents array Returns ------- parents : [ndarray

motions: list of Motion4Anim}