feat: pose estimation

[chmjkb]

Description

Adds a new pose estimation module to react-native-executorch, mirroring the object detection API surface.

• Native (C++): pose estimation runtime built on VisionModel, taking only modelSource + optional normMean / normStd. numKeypoints is derived from the model's output tensor
  shape — no keypointNames plumbed through the bridge.
• TS API: PoseEstimationModule (fromModelName / fromCustomModel) and usePoseEstimation hook. Return type is statically tied to keypointMapRuntime maps the native positional array into the named record at
  the boundary, in both forward and the runOnFrame worklet.
• Built-in models: yolo26n-pose (i will update more model sizes soon)
• Custom models: supported via fromCustomModel with user-supplied keypointMap; output type follows. Required model contract (input shape, optional normalization, three
  output tensors boxes / scores / keypoints) is documented.
• Docs: new usePoseEstimation and PoseEstimationModule pages following the object detection structure, plus VisionCamera integration entry.

Introduces a breaking change?

• Yes
• No

Type of change

• Bug fix (change which fixes an issue)
• New feature (change which adds functionality)
• Documentation update (improves or adds clarity to existing documentation)
• Other (chores, tests, code style improvements etc.)

Tested on

• iOS
• Android

Testing instructions

• Run native sanity tests and verify that they pass
• Run the computer vision example app, both for vision camera and static images.

Screenshots

Related issues

Checklist

• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have updated the documentation accordingly
• My changes generate no new warnings

Additional notes

[chmjkb]

Also, I am aware that there is significant overlap between this code and object detection. I will try to do something about it, but let's merge this first and let's do it in a separate PR.

tests are coming tho

[msluszniak]

Please rebase and resolve conflicts. Also, when I merged this one: 6acd575, I forgot to update README.md inside tests directory. But from now, you need to add all files used by a test suite to models_for_test (see attached commit). If you wouldn't mind, I send an update in README.md about this in your PR.

[msluszniak]

The lack of PoseEstimationTask.tsx blocks testing.

[benITo47]

+1, do we have an issue for that?

[chmjkb]

i'll create one

[msluszniak]

Please change names in demo app (see attached screens).

And I also tested vision camera integration and well. I'm not satisfied in 100% with the results. There are plenty of some random lines that go into the corner, etc. Here is a demo of the problems.

Screen_Recording_20260504_134403_computer-vision.mp4

And please add 'letterboxing' to cspell ignore, there is a warning about this one.

[chmjkb]

Please change names in demo app (see attached screens).

And I also tested vision camera integration and well. I'm not satisfied in 100% with the results. There are plenty of some random lines that go into the corner, etc. Here is a demo of the problems.

Screen_Recording_20260504_134403_computer-vision.mp4
And please add 'letterboxing' to cspell ignore, there is a warning about this one.

it was only happening for front camera (in portrait orientation), only on Android, pretty nasty to debug but the changes in FrameTransform should fix it.

[msluszniak]

And please add 'letterboxing' to cspell ignore, there is a warning about this one.

This one is still not addressed.

[chmjkb]

done cc @msluszniak

[msluszniak]

Ok, android works perfectly fine, now someone needs to test iOS and I think it's ready to ship :)

[benITo47]

I ran iOS app. The model downloads and runs correctly. Both on static images and VC feed. GJ.
Approved.
Godspeed