Architecture/Backbone for Human Segmentation?

[InternetMaster1]

Thanks for the amazing library!

I am looking to implement high-quality semantic segmentation on a mobile device for human segmentation. I am not just looking for portrait segmentation, but for full body segmentation (for single or multiple people).

What architecture/backbone would be a good choice for the task at hand? MobileNetV2, MobileNetV3, Unet, DeeplabV3+, PortraitNet, SINet.... There are so many, its confusing....

I want highest-acccuracy, rather than smallest or fastest, but it should not be so heavy that it becomes unsuitable for mobile.

Can you point me towards some comparison metrics/benchmarks?

What is your suggestion?

Thanks!