如何定义可以处理不同输入形状的图层？ 看起来像这样：
# dynamic_input_shape_conv_layer is a tf.keras layer
input_1 (None, 32, 32, 3) => dynamic_input_shape_conv_layer => (None, 32, 32, 6)
input_1 (None, 32, 32, 2) => dynamic_input_shape_conv_layer => (None, 32, 32, 6)
在TensorFlow 2.x中能否实现这一目标？