Any benchmarks comparing nanobbox?

[ogencoglu]

https://github.com/neka-nat/nanobbox

Btw I can not find fastbbox when I search in pypi.

[gerrylwk]

Hello, it's been a while. I'm trying to wrap up this project and publish to pypi, so it's not up there yet.

Tested with nanobind, and it's faster than Cython. You can try replacing the pyproject.toml with the pyproject.cython.toml.

I'm thinking if it makes sense to support both Cython and Nanobind on pypi. Feels a bit redundant if nanobind is faster

[gerrylwk]

#10 #9 #8
Comparing these 3 PRs, float32 seems to be the best option. Float16 implementation did not provide any significant gains, as ultimately the code still computes in float32. Maybe someone that knows nanobind better can take this up.

I suspect most people will only use the IoU function anyway. Proceeding with the float32 implementation

[gerrylwk]

Closed with #10 .
Package up on pypi: https://pypi.org/search/?q=fastbbox