python - tf.einsum matrix multiplication for different shape of tensors

Question

I have two tensors, a.shape = (?, 8, 448, 448, 64) representing (batch, head, height, weight, depth), b.shape = (895, 64), I want to get a new tensor c whose shape is (?, 8, 448, 448, 895). I used below code but failed with error: Subscripts have incorrect format

c = tf.einsum('bhlwd,nd?>bhlwn', a, b)

how to do it? except for tf.einsum, any other way to do it?

Answer 1

等待大神答复