Hugging Face's logo

qualcomm
/
HRNetPose

Keypoint Detection
PyTorch
LiteRT
android
Model card Files
xet
Community

HRNetPose: Optimized for Mobile Deployment

Perform accurate human pose estimation

HRNet performs pose estimation in high-resolution representations.

This model is an implementation of HRNetPose found here.

This repository provides scripts to run HRNetPose on Qualcomm® devices. More details on model performance across various devices, can be found here.

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
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
HRNetPose float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 14.342 ms 0 - 79 MB NPU HRNetPose.tflite
HRNetPose float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 14.101 ms 0 - 41 MB NPU HRNetPose.dlc
HRNetPose float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 3.64 ms 0 - 122 MB NPU HRNetPose.tflite
HRNetPose float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 4.773 ms 0 - 57 MB NPU HRNetPose.dlc
HRNetPose float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 2.668 ms 0 - 19 MB NPU HRNetPose.tflite
HRNetPose float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 2.607 ms 1 - 14 MB NPU HRNetPose.dlc
HRNetPose float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) ONNX 2.69 ms 0 - 125 MB NPU HRNetPose.onnx.zip
HRNetPose float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 4.348 ms 0 - 80 MB NPU HRNetPose.tflite
HRNetPose float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 4.265 ms 1 - 41 MB NPU HRNetPose.dlc
HRNetPose float SA7255P ADP Qualcomm® SA7255P TFLITE 14.342 ms 0 - 79 MB NPU HRNetPose.tflite
HRNetPose float SA7255P ADP Qualcomm® SA7255P QNN_DLC 14.101 ms 0 - 41 MB NPU HRNetPose.dlc
HRNetPose float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 2.618 ms 0 - 28 MB NPU HRNetPose.tflite
HRNetPose float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 2.596 ms 1 - 17 MB NPU HRNetPose.dlc
HRNetPose float SA8295P ADP Qualcomm® SA8295P TFLITE 4.514 ms 0 - 73 MB NPU HRNetPose.tflite
HRNetPose float SA8295P ADP Qualcomm® SA8295P QNN_DLC 4.496 ms 1 - 40 MB NPU HRNetPose.dlc
HRNetPose float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 2.623 ms 0 - 18 MB NPU HRNetPose.tflite
HRNetPose float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 2.612 ms 0 - 14 MB NPU HRNetPose.dlc
HRNetPose float SA8775P ADP Qualcomm® SA8775P TFLITE 4.348 ms 0 - 80 MB NPU HRNetPose.tflite
HRNetPose float SA8775P ADP Qualcomm® SA8775P QNN_DLC 4.265 ms 1 - 41 MB NPU HRNetPose.dlc
HRNetPose float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 1.893 ms 0 - 120 MB NPU HRNetPose.tflite
HRNetPose float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 1.982 ms 1 - 61 MB NPU HRNetPose.dlc
HRNetPose float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 2.008 ms 0 - 81 MB NPU HRNetPose.onnx.zip
HRNetPose float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile TFLITE 1.568 ms 0 - 83 MB NPU HRNetPose.tflite
HRNetPose float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile QNN_DLC 1.552 ms 1 - 50 MB NPU HRNetPose.dlc
HRNetPose float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile ONNX 1.687 ms 0 - 51 MB NPU HRNetPose.onnx.zip
HRNetPose float Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 2.866 ms 111 - 111 MB NPU HRNetPose.dlc
HRNetPose float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 2.725 ms 55 - 55 MB NPU HRNetPose.onnx.zip
HRNetPose w8a16 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 5.209 ms 0 - 67 MB NPU HRNetPose.dlc
HRNetPose w8a16 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 2.737 ms 0 - 93 MB NPU HRNetPose.dlc
HRNetPose w8a16 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 1.883 ms 0 - 23 MB NPU HRNetPose.dlc
HRNetPose w8a16 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 2.317 ms 0 - 67 MB NPU HRNetPose.dlc
HRNetPose w8a16 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN_DLC 8.776 ms 0 - 97 MB NPU HRNetPose.dlc
HRNetPose w8a16 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) ONNX 198.775 ms 18 - 46 MB CPU HRNetPose.onnx.zip
HRNetPose w8a16 RB5 (Proxy) Qualcomm® QCS8250 (Proxy) ONNX 210.372 ms 15 - 46 MB CPU HRNetPose.onnx.zip
HRNetPose w8a16 SA7255P ADP Qualcomm® SA7255P QNN_DLC 5.209 ms 0 - 67 MB NPU HRNetPose.dlc
HRNetPose w8a16 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 1.885 ms 0 - 16 MB NPU HRNetPose.dlc
HRNetPose w8a16 SA8295P ADP Qualcomm® SA8295P QNN_DLC 3.05 ms 0 - 71 MB NPU HRNetPose.dlc
HRNetPose w8a16 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 1.921 ms 0 - 15 MB NPU HRNetPose.dlc
HRNetPose w8a16 SA8775P ADP Qualcomm® SA8775P QNN_DLC 2.317 ms 0 - 67 MB NPU HRNetPose.dlc
HRNetPose w8a16 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 1.386 ms 0 - 92 MB NPU HRNetPose.dlc
HRNetPose w8a16 Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile QNN_DLC 1.019 ms 0 - 71 MB NPU HRNetPose.dlc
HRNetPose w8a16 Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile ONNX 101.666 ms 224 - 1921 MB NPU HRNetPose.onnx.zip
HRNetPose w8a16 Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 2.157 ms 138 - 138 MB NPU HRNetPose.dlc
HRNetPose w8a16 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 149.93 ms 226 - 226 MB NPU HRNetPose.onnx.zip
HRNetPose w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 2.6 ms 0 - 64 MB NPU HRNetPose.tflite
HRNetPose w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 2.83 ms 0 - 65 MB NPU HRNetPose.dlc
HRNetPose w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 1.186 ms 0 - 102 MB NPU HRNetPose.tflite
HRNetPose w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 1.684 ms 0 - 92 MB NPU HRNetPose.dlc
HRNetPose w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 0.952 ms 0 - 162 MB NPU HRNetPose.tflite
HRNetPose w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 1.171 ms 0 - 36 MB NPU HRNetPose.dlc
HRNetPose w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) ONNX 62.175 ms 108 - 363 MB NPU HRNetPose.onnx.zip
HRNetPose w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 1.298 ms 0 - 64 MB NPU HRNetPose.tflite
HRNetPose w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 1.47 ms 0 - 66 MB NPU HRNetPose.dlc
HRNetPose w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) TFLITE 3.689 ms 0 - 79 MB NPU HRNetPose.tflite
HRNetPose w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN_DLC 5.543 ms 0 - 91 MB NPU HRNetPose.dlc
HRNetPose w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) ONNX 142.09 ms 15 - 42 MB CPU HRNetPose.onnx.zip
HRNetPose w8a8 RB5 (Proxy) Qualcomm® QCS8250 (Proxy) TFLITE 17.59 ms 0 - 2 MB NPU HRNetPose.tflite
HRNetPose w8a8 RB5 (Proxy) Qualcomm® QCS8250 (Proxy) ONNX 128.583 ms 11 - 45 MB CPU HRNetPose.onnx.zip
HRNetPose w8a8 SA7255P ADP Qualcomm® SA7255P TFLITE 2.6 ms 0 - 64 MB NPU HRNetPose.tflite
HRNetPose w8a8 SA7255P ADP Qualcomm® SA7255P QNN_DLC 2.83 ms 0 - 65 MB NPU HRNetPose.dlc
HRNetPose w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 0.96 ms 0 - 165 MB NPU HRNetPose.tflite
HRNetPose w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 1.169 ms 0 - 47 MB NPU HRNetPose.dlc
HRNetPose w8a8 SA8295P ADP Qualcomm® SA8295P TFLITE 1.676 ms 0 - 68 MB NPU HRNetPose.tflite
HRNetPose w8a8 SA8295P ADP Qualcomm® SA8295P QNN_DLC 1.877 ms 0 - 72 MB NPU HRNetPose.dlc
HRNetPose w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 0.957 ms 0 - 163 MB NPU HRNetPose.tflite
HRNetPose w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 1.165 ms 0 - 82 MB NPU HRNetPose.dlc
HRNetPose w8a8 SA8775P ADP Qualcomm® SA8775P TFLITE 1.298 ms 0 - 64 MB NPU HRNetPose.tflite
HRNetPose w8a8 SA8775P ADP Qualcomm® SA8775P QNN_DLC 1.47 ms 0 - 66 MB NPU HRNetPose.dlc
HRNetPose w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 0.704 ms 0 - 98 MB NPU HRNetPose.tflite
HRNetPose w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 0.82 ms 0 - 94 MB NPU HRNetPose.dlc
HRNetPose w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 47.296 ms 122 - 2091 MB NPU HRNetPose.onnx.zip
HRNetPose w8a8 Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile TFLITE 0.574 ms 0 - 71 MB NPU HRNetPose.tflite
HRNetPose w8a8 Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile QNN_DLC 0.631 ms 0 - 74 MB NPU HRNetPose.dlc
HRNetPose w8a8 Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile ONNX 43.047 ms 94 - 869 MB NPU HRNetPose.onnx.zip
HRNetPose w8a8 Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 1.305 ms 134 - 134 MB NPU HRNetPose.dlc
HRNetPose w8a8 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 60.036 ms 116 - 116 MB NPU HRNetPose.onnx.zip

Installation

Install the package via pip:

pip install "qai-hub-models[hrnet-pose]" torch==2.4.1 --trusted-host download.openmmlab.com -f https://download.openmmlab.com/mmcv/dist/cpu/torch2.4/index.html -f https://qaihub-public-python-wheels.s3.us-west-2.amazonaws.com/index.html

Configure Qualcomm® AI Hub to run this model on a cloud-hosted device

Sign-in to Qualcomm® AI Hub with your Qualcomm® ID. Once signed in navigate to Account -> Settings -> API Token.

With this API token, you can configure your client to run models on the cloud hosted devices.

qai-hub configure --api_token API_TOKEN

Navigate to docs for more information.

Demo off target

The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input.

python -m qai_hub_models.models.hrnet_pose.demo

The above demo runs a reference implementation of pre-processing, model inference, and post processing.

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.hrnet_pose.demo

Run model on a cloud-hosted device

In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following:

  • Performance check on-device on a cloud-hosted device
  • Downloads compiled assets that can be deployed on-device for Android.
  • Accuracy check between PyTorch and on-device outputs.
python -m qai_hub_models.models.hrnet_pose.export

How does this work?

This export script leverages Qualcomm® AI Hub to optimize, validate, and deploy this model on-device. Lets go through each step below in detail:

Step 1: Compile model for on-device deployment

To compile a PyTorch model for on-device deployment, we first trace the model in memory using the jit.trace and then call the submit_compile_job API.

import torch

import qai_hub as hub
from qai_hub_models.models.hrnet_pose import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S25")

# Trace model
input_shape = torch_model.get_input_spec()
sample_inputs = torch_model.sample_inputs()

pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])

# Compile model on a specific device
compile_job = hub.submit_compile_job(
    model=pt_model,
    device=device,
    input_specs=torch_model.get_input_spec(),
)

# Get target model to run on-device
target_model = compile_job.get_target_model()

Step 2: Performance profiling on cloud-hosted device

After compiling models from step 1. Models can be profiled model on-device using the target_model. Note that this scripts runs the model on a device automatically provisioned in the cloud. Once the job is submitted, you can navigate to a provided job URL to view a variety of on-device performance metrics.

profile_job = hub.submit_profile_job(
    model=target_model,
    device=device,
)

Step 3: Verify on-device accuracy

To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device.

input_data = torch_model.sample_inputs()
inference_job = hub.submit_inference_job(
    model=target_model,
    device=device,
    inputs=input_data,
)
    on_device_output = inference_job.download_output_data()

With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output.

Note: This on-device profiling and inference requires access to Qualcomm® AI Hub. Sign up for access.

Run demo on a cloud-hosted device

You can also run the demo on-device.

python -m qai_hub_models.models.hrnet_pose.demo --eval-mode on-device

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.hrnet_pose.demo -- --eval-mode on-device

Deploying compiled model to Android

The models can be deployed using multiple runtimes:

  • TensorFlow Lite (.tflite export): This tutorial provides a guide to deploy the .tflite model in an Android application.

  • QNN (.so export ): This sample app provides instructions on how to use the .so shared library in an Android application.

View on Qualcomm® AI Hub

Get more details on HRNetPose's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

  • The license for the original implementation of HRNetPose can be found here.
  • The license for the compiled assets for on-device deployment can be found here

References

Community

Downloads last month
1,186
Inference Providers NEW
Keypoint Detection
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support