Computes multi-headed attention. :param k: keys [B, M, D] with M being the sentence length. :param v: values [B, M, D]

Embed inputs. Args: emb (FloatTensor): Sequence of word vectors ``(seq_len, batch_size, self.dim)``

Forward pass for a single transformer encoder layer. First applies layer norm, then self attention, then dropout with residua

Forward pass of a single Transformer decoder layer. :param x: inputs :param memory: source representations :param sr

Perform lookup for input `x` in the embedding table. :param mask: token masks :param x: index in the vocabulary :ret

:param mask: frame masks :param x: input frame features :return: embedded representation for `x`

Returns the mean of the mixture

Compute 2D reprojection loss on the keypoints. The loss is weighted by the confidence. The available keypoints are different for each dat

Convert rotations given as rotation matrices to axis/angle. Args: matrix: Rotation matrices as tensor of shape (..., 3, 3). Ret

Convert rotations given as rotation matrices to axis/angle. Args: matrix: Rotation matrices as tensor of shape (..., 3, 3). Ret

Convert rotations given as rotation matrices to axis/angle. Args: matrix: Rotation matrices as tensor of shape (..., 3, 3). Ret

Convert rotations given as rotation matrices to Euler angles in radians. Args: matrix: Rotation matrices as tensor of shape (..., 3,

Convert rotations given as rotation matrices to Euler angles in radians. Args: matrix: Rotation matrices as tensor of shape (..., 3,

Convert rotations given as rotation matrices to Euler angles in radians. Args: matrix: Rotation matrices as tensor of shape (..., 3,

Converts rotation matrices to 6D rotation representation by Zhou et al. [1] by dropping the last row. Note that 6D representation is not uniq