Maybe the dropblock is not right?

[Jamesgender]

Thanks for your nice work!
But I have a question about dropblock.The original paper writes "We only sample mask from shaded green region in which each sampled entry can expanded to a mask fully contained inside the feature map"

So the mask sample point will smaller than the input and then expanded to the input size. In your code, it seems like the sample points get from the input size.
Am I right? Thanks again for your GREAT work!