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

---

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

# HRNetPose: Optimized for Qualcomm Devices

HRNet performs pose estimation in high-resolution representations.

This is based on the implementation of HRNetPose found [here](https://github.com/leoxiaobin/deep-high-resolution-net.pytorch).
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/hrnet_pose) 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 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/hrnet_pose/releases/v0.46.0/hrnet_pose-onnx-float.zip)
| ONNX | w8a16 | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/hrnet_pose/releases/v0.46.0/hrnet_pose-onnx-w8a16.zip)
| QNN_DLC | float | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/hrnet_pose/releases/v0.46.0/hrnet_pose-qnn_dlc-float.zip)
| QNN_DLC | w8a16 | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/hrnet_pose/releases/v0.46.0/hrnet_pose-qnn_dlc-w8a16.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/hrnet_pose/releases/v0.46.0/hrnet_pose-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/hrnet_pose/releases/v0.46.0/hrnet_pose-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/hrnet_pose/releases/v0.46.0/hrnet_pose-tflite-w8a8.zip)

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


### 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/hrnet_pose) 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 [HRNetPose on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/hrnet_pose) for usage instructions.

## Model Details

**Model Type:** Model_use_case.pose_estimation

**Model Stats:**
- Model checkpoint: hrnet_posenet_FP32_state_dict
- Input resolution: 256x192
- Number of parameters: 28.5M
- Model size (float): 109 MB
- Model size (w8a8): 28.1 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| HRNetPose | ONNX | float | Snapdragon® X Elite | 2.663 ms | 55 - 55 MB | NPU
| HRNetPose | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 2.073 ms | 0 - 197 MB | NPU
| HRNetPose | ONNX | float | Qualcomm® QCS8550 (Proxy) | 2.802 ms | 1 - 3 MB | NPU
| HRNetPose | ONNX | float | Qualcomm® QCS9075 | 4.107 ms | 0 - 4 MB | NPU
| HRNetPose | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.748 ms | 0 - 142 MB | NPU
| HRNetPose | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.423 ms | 0 - 143 MB | NPU
| HRNetPose | ONNX | w8a16 | Snapdragon® X Elite | 1.978 ms | 28 - 28 MB | NPU
| HRNetPose | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.511 ms | 0 - 258 MB | NPU
| HRNetPose | ONNX | w8a16 | Qualcomm® QCS6490 | 486.81 ms | 32 - 41 MB | CPU
| HRNetPose | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 2.014 ms | 0 - 268 MB | NPU
| HRNetPose | ONNX | w8a16 | Qualcomm® QCS9075 | 2.297 ms | 0 - 3 MB | NPU
| HRNetPose | ONNX | w8a16 | Qualcomm® QCM6690 | 214.195 ms | 30 - 48 MB | CPU
| HRNetPose | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 1.232 ms | 0 - 188 MB | NPU
| HRNetPose | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 209.371 ms | 26 - 39 MB | CPU
| HRNetPose | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.996 ms | 0 - 189 MB | NPU
| HRNetPose | QNN_DLC | float | Snapdragon® X Elite | 2.876 ms | 1 - 1 MB | NPU
| HRNetPose | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 2.006 ms | 0 - 121 MB | NPU
| HRNetPose | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 14.33 ms | 1 - 75 MB | NPU
| HRNetPose | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 2.677 ms | 1 - 87 MB | NPU
| HRNetPose | QNN_DLC | float | Qualcomm® SA8775P | 18.856 ms | 1 - 75 MB | NPU
| HRNetPose | QNN_DLC | float | Qualcomm® QCS9075 | 4.118 ms | 1 - 3 MB | NPU
| HRNetPose | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 4.961 ms | 0 - 105 MB | NPU
| HRNetPose | QNN_DLC | float | Qualcomm® SA7255P | 14.33 ms | 1 - 75 MB | NPU
| HRNetPose | QNN_DLC | float | Qualcomm® SA8295P | 4.517 ms | 0 - 67 MB | NPU
| HRNetPose | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.586 ms | 1 - 77 MB | NPU
| HRNetPose | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.267 ms | 0 - 78 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Snapdragon® X Elite | 2.162 ms | 0 - 0 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.394 ms | 0 - 151 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Qualcomm® QCS6490 | 6.753 ms | 0 - 2 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 5.247 ms | 0 - 98 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 1.918 ms | 0 - 2 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Qualcomm® SA8775P | 2.261 ms | 0 - 100 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 2.196 ms | 2 - 4 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 21.984 ms | 0 - 217 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 2.609 ms | 0 - 152 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Qualcomm® SA7255P | 5.247 ms | 0 - 98 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Qualcomm® SA8295P | 3.086 ms | 0 - 96 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 1.018 ms | 0 - 99 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 2.557 ms | 0 - 101 MB | NPU
| HRNetPose | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.812 ms | 0 - 100 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Snapdragon® X Elite | 1.269 ms | 0 - 0 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.829 ms | 0 - 133 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 3.664 ms | 2 - 4 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 2.869 ms | 0 - 88 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 1.124 ms | 0 - 4 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Qualcomm® SA8775P | 5.323 ms | 0 - 88 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 1.309 ms | 0 - 2 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 10.276 ms | 0 - 209 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 1.714 ms | 0 - 135 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Qualcomm® SA7255P | 2.869 ms | 0 - 88 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Qualcomm® SA8295P | 1.893 ms | 0 - 87 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.642 ms | 0 - 92 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.507 ms | 0 - 88 MB | NPU
| HRNetPose | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.564 ms | 0 - 92 MB | NPU
| HRNetPose | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 2.028 ms | 0 - 195 MB | NPU
| HRNetPose | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 14.366 ms | 0 - 120 MB | NPU
| HRNetPose | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 2.675 ms | 0 - 3 MB | NPU
| HRNetPose | TFLITE | float | Qualcomm® SA8775P | 4.37 ms | 0 - 119 MB | NPU
| HRNetPose | TFLITE | float | Qualcomm® QCS9075 | 4.13 ms | 0 - 58 MB | NPU
| HRNetPose | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 4.964 ms | 0 - 180 MB | NPU
| HRNetPose | TFLITE | float | Qualcomm® SA7255P | 14.366 ms | 0 - 120 MB | NPU
| HRNetPose | TFLITE | float | Qualcomm® SA8295P | 4.555 ms | 0 - 107 MB | NPU
| HRNetPose | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.583 ms | 0 - 120 MB | NPU
| HRNetPose | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.266 ms | 0 - 121 MB | NPU
| HRNetPose | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.701 ms | 0 - 143 MB | NPU
| HRNetPose | TFLITE | w8a8 | Qualcomm® QCS6490 | 3.334 ms | 0 - 30 MB | NPU
| HRNetPose | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 2.605 ms | 0 - 88 MB | NPU
| HRNetPose | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.966 ms | 0 - 3 MB | NPU
| HRNetPose | TFLITE | w8a8 | Qualcomm® SA8775P | 1.301 ms | 0 - 92 MB | NPU
| HRNetPose | TFLITE | w8a8 | Qualcomm® QCS9075 | 1.097 ms | 0 - 30 MB | NPU
| HRNetPose | TFLITE | w8a8 | Qualcomm® QCM6690 | 9.757 ms | 0 - 200 MB | NPU
| HRNetPose | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 1.485 ms | 0 - 145 MB | NPU
| HRNetPose | TFLITE | w8a8 | Qualcomm® SA7255P | 2.605 ms | 0 - 88 MB | NPU
| HRNetPose | TFLITE | w8a8 | Qualcomm® SA8295P | 1.723 ms | 0 - 85 MB | NPU
| HRNetPose | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.583 ms | 0 - 91 MB | NPU
| HRNetPose | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.341 ms | 0 - 86 MB | NPU
| HRNetPose | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.518 ms | 0 - 92 MB | NPU

## License
* The license for the original implementation of HRNetPose can be found
  [here](https://github.com/leoxiaobin/deep-high-resolution-net.pytorch/blob/master/LICENSE).

## References
* [Deep High-Resolution Representation Learning for Human Pose Estimation](https://arxiv.org/abs/1902.09212)
* [Source Model Implementation](https://github.com/leoxiaobin/deep-high-resolution-net.pytorch)

## 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).