README.md

---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: keypoint-detection

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/rtmpose_body2d/web-assets/model_demo.png)

# RTMPose-Body2d: Optimized for Qualcomm Devices

RTMPose is a machine learning model that detects human pose and returns a location and confidence for each of 133 joints.

This is based on the implementation of RTMPose-Body2d found [here](https://github.com/open-mmlab/mmpose/tree/main/projects/rtmpose).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/rtmpose_body2d) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.

## Getting Started
There are two ways to deploy this model on your device:

### Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

| Runtime | Precision | Chipset | SDK Versions | Download |
|---|---|---|---|---|
| ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/rtmpose_body2d/releases/v0.47.0/rtmpose_body2d-onnx-float.zip)
| ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/rtmpose_body2d/releases/v0.47.0/rtmpose_body2d-onnx-w8a16.zip)
| QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/rtmpose_body2d/releases/v0.47.0/rtmpose_body2d-qnn_dlc-float.zip)
| QNN_DLC | w8a16 | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/rtmpose_body2d/releases/v0.47.0/rtmpose_body2d-qnn_dlc-w8a16.zip)
| TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/rtmpose_body2d/releases/v0.47.0/rtmpose_body2d-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[RTMPose-Body2d on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/rtmpose_body2d)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/rtmpose_body2d) Python library to compile and export the model with your own:
- Custom weights (e.g., fine-tuned checkpoints)
- Custom input shapes
- Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for [RTMPose-Body2d on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/rtmpose_body2d) for usage instructions.

## Model Details

**Model Type:** Model_use_case.pose_estimation

**Model Stats:**
- Input resolution: 256x192
- Number of parameters: 17.9M
- Model size (float): 68.5 MB
- Model size (w8a16): 18.2 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| RTMPose-Body2d | ONNX | float | Snapdragon® X Elite | 1.873 ms | 37 - 37 MB | NPU
| RTMPose-Body2d | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 1.338 ms | 0 - 55 MB | NPU
| RTMPose-Body2d | ONNX | float | Qualcomm® QCS8550 (Proxy) | 1.765 ms | 0 - 41 MB | NPU
| RTMPose-Body2d | ONNX | float | Qualcomm® QCS9075 | 2.437 ms | 0 - 4 MB | NPU
| RTMPose-Body2d | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.091 ms | 0 - 35 MB | NPU
| RTMPose-Body2d | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.921 ms | 0 - 38 MB | NPU
| RTMPose-Body2d | ONNX | float | Snapdragon® X2 Elite | 0.933 ms | 37 - 37 MB | NPU
| RTMPose-Body2d | ONNX | w8a16 | Snapdragon® X Elite | 1.91 ms | 19 - 19 MB | NPU
| RTMPose-Body2d | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.169 ms | 0 - 87 MB | NPU
| RTMPose-Body2d | ONNX | w8a16 | Qualcomm® QCS6490 | 175.399 ms | 53 - 56 MB | CPU
| RTMPose-Body2d | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 1.71 ms | 0 - 36 MB | NPU
| RTMPose-Body2d | ONNX | w8a16 | Qualcomm® QCS9075 | 1.904 ms | 0 - 3 MB | NPU
| RTMPose-Body2d | ONNX | w8a16 | Qualcomm® QCM6690 | 86.92 ms | 48 - 57 MB | CPU
| RTMPose-Body2d | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.868 ms | 0 - 61 MB | NPU
| RTMPose-Body2d | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 82.722 ms | 47 - 55 MB | CPU
| RTMPose-Body2d | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.736 ms | 0 - 61 MB | NPU
| RTMPose-Body2d | ONNX | w8a16 | Snapdragon® X2 Elite | 0.77 ms | 19 - 19 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Snapdragon® X Elite | 1.84 ms | 1 - 1 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 1.331 ms | 0 - 57 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 7.51 ms | 1 - 33 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 1.738 ms | 1 - 2 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Qualcomm® SA8775P | 2.429 ms | 1 - 36 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Qualcomm® QCS9075 | 2.433 ms | 1 - 3 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 3.514 ms | 0 - 62 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Qualcomm® SA7255P | 7.51 ms | 1 - 33 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Qualcomm® SA8295P | 3.533 ms | 0 - 35 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.074 ms | 1 - 34 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.905 ms | 1 - 36 MB | NPU
| RTMPose-Body2d | QNN_DLC | float | Snapdragon® X2 Elite | 1.087 ms | 1 - 1 MB | NPU
| RTMPose-Body2d | QNN_DLC | w8a16 | Snapdragon® X Elite | 1.951 ms | 0 - 0 MB | NPU
| RTMPose-Body2d | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.18 ms | 0 - 75 MB | NPU
| RTMPose-Body2d | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 3.835 ms | 0 - 50 MB | NPU
| RTMPose-Body2d | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 1.732 ms | 0 - 2 MB | NPU
| RTMPose-Body2d | QNN_DLC | w8a16 | Qualcomm® SA8775P | 2.04 ms | 0 - 52 MB | NPU
| RTMPose-Body2d | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 1.904 ms | 0 - 2 MB | NPU
| RTMPose-Body2d | QNN_DLC | w8a16 | Qualcomm® SA7255P | 3.835 ms | 0 - 50 MB | NPU
| RTMPose-Body2d | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.89 ms | 0 - 53 MB | NPU
| RTMPose-Body2d | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.753 ms | 0 - 51 MB | NPU
| RTMPose-Body2d | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 0.974 ms | 0 - 0 MB | NPU
| RTMPose-Body2d | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.327 ms | 0 - 91 MB | NPU
| RTMPose-Body2d | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 7.468 ms | 0 - 49 MB | NPU
| RTMPose-Body2d | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 1.755 ms | 0 - 2 MB | NPU
| RTMPose-Body2d | TFLITE | float | Qualcomm® SA8775P | 2.455 ms | 0 - 51 MB | NPU
| RTMPose-Body2d | TFLITE | float | Qualcomm® QCS9075 | 2.425 ms | 0 - 40 MB | NPU
| RTMPose-Body2d | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 3.556 ms | 0 - 93 MB | NPU
| RTMPose-Body2d | TFLITE | float | Qualcomm® SA7255P | 7.468 ms | 0 - 49 MB | NPU
| RTMPose-Body2d | TFLITE | float | Qualcomm® SA8295P | 3.553 ms | 0 - 51 MB | NPU
| RTMPose-Body2d | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.078 ms | 0 - 52 MB | NPU
| RTMPose-Body2d | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.901 ms | 0 - 49 MB | NPU

## License
* The license for the original implementation of RTMPose-Body2d can be found
  [here](https://github.com/open-mmlab/mmpose/blob/main/LICENSE).

## References
* [RTMPose: Real-Time Multi-Person Pose Estimation based on MMPose](https://arxiv.org/abs/2303.07399)
* [Source Model Implementation](https://github.com/open-mmlab/mmpose/tree/main/projects/rtmpose)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).