pytorch bug?

[lzmchina]

I want to use adaptive_avg_pool2d function to transform input tensor to output tensor:

output = adaptive_avg_pool2d(input, new LongArrayRef(new long[] { 7, 7 }, 2));

The input variable's shape is [64, 512, 7, 7], but the output's shape is [64, 512, 0, 7].

The correct output shape should be [64, 512, 7, 7]

[saudet]

You're probably not calling the function you want to call. Try to call the one with AdaptiveAvgPool2dOptions as parameter.

[lzmchina]

You're probably not calling the function you want to call. Try to call the one with AdaptiveAvgPool2dOptions as parameter.

I have tried it at the first, but I still got a wrong shape tensor.

[saudet]

If you have a working example in C++, please provide it. It should be easy to translate it to Java.

[saudet]

Ah, I see what's wrong. We need an array of LongOptional for this. I've pushed a fix in commit 94268bb, and it's a bit awkward, but it works this way:

LongOptionalVector size = new LongOptionalVector(new LongOptional(7), new LongOptional(7));
Tensor output = adaptive_avg_pool2d(input, size.get(0));

Thanks for reporting!

[saudet]

The fix for this has been released with version 1.5.8! Enjoy