Person-Foot-Detection-Quantized: Optimized for Mobile Deployment

Multi-task Human detector

Real-time multiple person detection with accurate feet localization optimized for mobile and edge.

This model is an implementation of Person-Foot-Detection-Quantized found here.

This repository provides scripts to run Person-Foot-Detection-Quantized on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

Model Type: Object detection
Model Stats:
- Model checkpoint: SA-e30_finetune50.pth
- Inference latency: RealTime
- Input resolution: 640x480
- Number of output classes: 2
- Number of parameters: 2.53M
- Model size: 9.69 MB

Model	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Precision	Primary Compute Unit	Target Model
Person-Foot-Detection-Quantized	Samsung Galaxy S23	Snapdragon® 8 Gen 2	TFLITE	1.144 ms	0 - 5 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	Samsung Galaxy S23	Snapdragon® 8 Gen 2	QNN	1.288 ms	0 - 5 MB	INT8	NPU	Person-Foot-Detection-Quantized.so
Person-Foot-Detection-Quantized	Samsung Galaxy S23	Snapdragon® 8 Gen 2	ONNX	1.661 ms	0 - 4 MB	INT8	NPU	Person-Foot-Detection-Quantized.onnx
Person-Foot-Detection-Quantized	Samsung Galaxy S24	Snapdragon® 8 Gen 3	TFLITE	0.797 ms	0 - 29 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	Samsung Galaxy S24	Snapdragon® 8 Gen 3	QNN	0.867 ms	1 - 24 MB	INT8	NPU	Person-Foot-Detection-Quantized.so
Person-Foot-Detection-Quantized	Samsung Galaxy S24	Snapdragon® 8 Gen 3	ONNX	1.167 ms	0 - 58 MB	INT8	NPU	Person-Foot-Detection-Quantized.onnx
Person-Foot-Detection-Quantized	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	TFLITE	0.774 ms	0 - 21 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	QNN	0.84 ms	0 - 22 MB	INT8	NPU	Use Export Script
Person-Foot-Detection-Quantized	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	ONNX	1.126 ms	0 - 40 MB	INT8	NPU	Person-Foot-Detection-Quantized.onnx
Person-Foot-Detection-Quantized	RB3 Gen 2 (Proxy)	QCS6490 Proxy	TFLITE	5.661 ms	1 - 30 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	RB3 Gen 2 (Proxy)	QCS6490 Proxy	QNN	6.809 ms	1 - 7 MB	INT8	NPU	Use Export Script
Person-Foot-Detection-Quantized	RB5 (Proxy)	QCS8250 Proxy	TFLITE	26.3 ms	1 - 11 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	QCS8550 (Proxy)	QCS8550 Proxy	TFLITE	1.145 ms	0 - 6 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	QCS8550 (Proxy)	QCS8550 Proxy	QNN	1.2 ms	1 - 2 MB	INT8	NPU	Use Export Script
Person-Foot-Detection-Quantized	SA7255P ADP	SA7255P	TFLITE	19.55 ms	1 - 26 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	SA7255P ADP	SA7255P	QNN	19.789 ms	1 - 11 MB	INT8	NPU	Use Export Script
Person-Foot-Detection-Quantized	SA8255 (Proxy)	SA8255P Proxy	TFLITE	1.149 ms	0 - 101 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	SA8255 (Proxy)	SA8255P Proxy	QNN	1.247 ms	1 - 3 MB	INT8	NPU	Use Export Script
Person-Foot-Detection-Quantized	SA8295P ADP	SA8295P	TFLITE	2.25 ms	0 - 20 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	SA8295P ADP	SA8295P	QNN	2.32 ms	1 - 7 MB	INT8	NPU	Use Export Script
Person-Foot-Detection-Quantized	SA8650 (Proxy)	SA8650P Proxy	TFLITE	1.155 ms	0 - 70 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	SA8650 (Proxy)	SA8650P Proxy	QNN	1.237 ms	0 - 1 MB	INT8	NPU	Use Export Script
Person-Foot-Detection-Quantized	SA8775P ADP	SA8775P	TFLITE	1.884 ms	0 - 25 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	SA8775P ADP	SA8775P	QNN	2.075 ms	1 - 7 MB	INT8	NPU	Use Export Script
Person-Foot-Detection-Quantized	QCS8450 (Proxy)	QCS8450 Proxy	TFLITE	1.418 ms	0 - 26 MB	INT8	NPU	Person-Foot-Detection-Quantized.tflite
Person-Foot-Detection-Quantized	QCS8450 (Proxy)	QCS8450 Proxy	QNN	1.553 ms	1 - 28 MB	INT8	NPU	Use Export Script
Person-Foot-Detection-Quantized	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN	1.452 ms	1 - 1 MB	INT8	NPU	Use Export Script
Person-Foot-Detection-Quantized	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	1.776 ms	8 - 8 MB	INT8	NPU	Person-Foot-Detection-Quantized.onnx

Installation

This model can be installed as a Python package via pip.

pip install "qai-hub-models[foot_track_net_quantized]"

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.foot_track_net_quantized.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.foot_track_net_quantized.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.foot_track_net_quantized.export

Profiling Results
------------------------------------------------------------
Person-Foot-Detection-Quantized
Device                          : Samsung Galaxy S23 (13)
Runtime                         : TFLITE                 
Estimated inference time (ms)   : 1.1                    
Estimated peak memory usage (MB): [0, 5]                 
Total # Ops                     : 146                    
Compute Unit(s)                 : NPU (146 ops)

Run demo on a cloud-hosted device

You can also run the demo on-device.

python -m qai_hub_models.models.foot_track_net_quantized.demo --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.foot_track_net_quantized.demo -- --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 Person-Foot-Detection-Quantized's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of Person-Foot-Detection-Quantized can be found here.
The license for the compiled assets for on-device deployment can be found here

References

Community

Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
For questions or feedback please reach out to us.