ISSAC AprilTag detector worse than open source one

I am comparing 10k images between the original CPU version and the one from ISSAC and I noticed that the ISSAC version is at least 12% worse compared to the CPU version:

Original CPU version from UMich:

ISSAC GPU version:

Sample images can be provided through emails.

Could NVIDIA explains the implementation details of this ISSAC AprilTag detector?
I though only the image processing part has been accelerated by GPU and the detection accuracy should be at least as good as the open source one.

Hello dk1900, NVIDIA’s implementation uses GPU. GPU based AprilTag is a different algorithm than the CPU open source algorithm. It has much faster performance (~20 times) and also a little lower detection rate (~10%) esp when the AprilTag is at the edge of a wide field of view. For best detection rate we recommend that the AprilTag be within the 90 degree FOV of the camera.

Hello. I can not find any links for this headers and library, please say where you get it? :) Can I got there other headers?
I want direct usage without node system…