Deconvolution can easily have “uneven overlap,”（不均匀重叠）putting more of the metaphorical paint in some places than others. In particular, deconvolution has uneven overlap when the kernel size (the output window size) is not divisible by the stride (the spacing between points on the top). While the network could, in principle, carefully learn weights to avoid this — as we’ll discuss in more detail later — in practice neural networks struggle to avoid it completely.