Adapt tensor output_size for conv2d_transpose and depthwise_conv2d_transpose