feat: speed up box iou batch using 2d in-place ops instead of 3d

[AnonymDevOSS]

Description

I focused on improving the performance of the IoU calculation between bounding boxes.

This change adds a new function box_iou_batch_alt, which performs the same computation as box_iou_batch but runs faster and uses less memory.

The optimization uses in-place NumPy operations and avoids creating temporary arrays.
Results are identical to the original implementation within a small numerical tolerance.

No new dependencies were added.

Type of change

Performance improvement (non-breaking change)

How has this change been tested, please provide a testcase or example of how you tested the change?

I compared the outputs of both functions (box_iou_batch and box_iou_batch_alt) using random test boxes.
A simple pytest test checks that the results are equal within rtol=1e-6 and atol=1e-6.
Both functions return the same values for IoU and IoS modes.
Benchmarks show that the new version is around 2–5× faster on typical box sizes.

I also add a benchmark.

(.venv) wednesday:benchmark rafelbennasar$ python benchmark_iou.py 
Benchmarking 2 functions on sizes=[(512, 1024)], 5 runs × 5 iterations each

- box_iou_batch  (512×1024)
[benchmark_iou.py](https://github.com/user-attachments/files/23167441/benchmark_iou.py)

  Run 01:   0.241 s
  Run 02:   0.232 s
  Run 03:   0.522 s
  Run 04:   0.246 s
  Run 05:   0.207 s
  Average:   0.290 s

- box_iou_batch_alt  (512×1024)
  Run 01:   0.060 s
  Run 02:   0.064 s
  Run 03:   0.066 s
  Run 04:   0.065 s
  Run 05:   0.066 s
  Average:   0.064 s


= Summary (5 runs × 5 iterations each):
box_iou_batch_alt           avg=  0.064s  stdev= 0.002s  min= 0.060s  max= 0.066s
box_iou_batch               avg=  0.290s  stdev= 0.131s  min= 0.207s  max= 0.522s

[CLAassistant]

All committers have signed the CLA.

[SkalskiP]

Hi @AnonymDevOSS 👋🏻 thanks a lot for the contribution!

Please accept the CLA license so we can move forward.

[AnonymDevOSS]

I’m a little confused because I believe I’ve already accepted the license agreement.

[SkalskiP]

Looks like you used different account to commit the code.

[AnonymDevOSS]

Oops. Thanks for the heads up! Fixed.

[SkalskiP]

Let's rename it to box_iou_batch and remove old implementation of this function.

[SkalskiP]

In supervision/test/test_utils.py, there are already two functions, mock_detections and mock_key_points. It seems we are duplicating some logic from mock_detections. Try to unify them.

If that’s not possible, at least move generate_boxes to supervision/test/test_utils.py and make its arguments and naming conventions consistent with the existing functions.

One quick improvement would be to replace the separate W and H arguments with a single resolution_wh argument, which we use throughout the codebase.

[SkalskiP]

As we will rename box_iou_batch_alt to box_iou_batch that test no longer will be necessary.

[AnonymDevOSS]

Added additional test cases rather than using the trivial approach of comparing against the original implementation.

[SkalskiP]

Hi @AnonymDevOSS 👋🏻 Sorry for the delay, I’ve been a bit busy with other tasks at Roboflow. I left a few comments on the PR. The supervision-0.27.0 release is coming up, so if you manage to apply the changes soon, this PR could be included in that release.