PyTorch AdaptiveAvgPool2dImpl constructor

[haifengl]

AdaptiveAvgPool2dImpl constructor takes a parameter output_size. It is the target output size of the image of the form H x W.
Can be a tuple (H, W) or a single H for a square image H x H. But currently the constructor takes only LongOptional as input. This will cause the second parameter be undefined number during runtime and cause errors.

[HGuillemet]

LongOptional is a subclass of Pointer. Like any Pointer, it can point to a single item or an array.
In the case of AdaptiveAvgPool2dImpl you must supply an array of 2 LongOptional.
The problem here is that, for LongOptional, the usual array constructor that takes the array length as argument is shadowed by the constructor that takes the optional long.
You should be able to use a vector instead:

AdaptiveAvgPool2dImpl aap = new AdaptiveAvgPool2dImpl(new LongOptionalVector(new LongOptional(10), new LongOptional(12)).front());

Can you give it a try ?

[haifengl]

It seems work. I will further check if the computation result is correct. Thanks! I would suggest changing the API to LongPointer instead of LongOptional. In fact, Conv2dImpl's kernel parameter is similar and it takes LongPointer. It is clear and less misleading.