tfa.image.rotate returns <KerasTensor: shape=(None, None, None, 3) dtype=float32 (created by layer 'lambda_6')>

[Isaac Gerg]

input_squints

<KerasTensor: shape=(None, 512, 512, 3) dtype=float32 (created by layer 'tf.math.truediv_8')>

input_rotation_layer

<KerasTensor: shape=(None, 1) dtype=float32 (created by layer 'input_2')>

tf.keras.layers.Lambda(lambda x:tfa.image.rotate(x[0], x[1]))([input_squints, input_rotation_layer[:,0]])

I would expect the line of code above to return a tensor of shape (None, 512,512,3).  Why does it not do this?

[Isaac Gerg]

I get a tensor with shape (None, None, None, 3)