---
library_name: onnxruntime
pipeline_tag: keypoint-detection
tags:
- wardy
- on-device-ai
- jetson
- pose-estimation
license: apache-2.0
---

# Wardy M-03 pose estimation

This repository pins the RTMPose-T ONNX artifact used by Wardy M-03 for COCO-17 body keypoint estimation.

## Artifact

| Runtime version | File |
|---|---|
| `openmmlab-rtmpose-t-20230504` | `model.onnx` |

`manifest.json` contains the SHA-256 digest verified by the Wardy model manager.

## Provenance and license

The model artifact was published by OpenMMLab as part of RTMPose and is redistributed under Apache-2.0. Wardy does not claim original model authorship.

## Runtime boundary

Wardy downloads the pinned ONNX artifact, verifies its digest, and runs pose inference locally. Device-specific TensorRT engines are generated on the target Jetson and are not distributed here.

## Limitations

Pose accuracy depends on person visibility, scale, lighting, occlusion, and the preceding person detector. Pose output alone does not determine a fall.