MediaPipe-Face-Detection: Optimized for Mobile Deployment

Detect faces and locate facial features in real-time video and image streams

Designed for sub-millisecond processing, this model predicts bounding boxes and pose skeletons (left eye, right eye, nose tip, mouth, left eye tragion, and right eye tragion) of faces in an image.

This model is an implementation of MediaPipe-Face-Detection found here.

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

Model Details

Model Type: Object detection
Model Stats:
- Input resolution: 256x256
- Number of parameters (MediaPipeFaceDetector): 135K
- Model size (MediaPipeFaceDetector): 565 KB
- Number of parameters (MediaPipeFaceLandmarkDetector): 603K
- Model size (MediaPipeFaceLandmarkDetector): 2.34 MB
- Number of output classes: 6

Model	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Precision	Primary Compute Unit	Target Model
FaceDetector	Samsung Galaxy S23	Snapdragon® 8 Gen 2	TFLITE	0.553 ms	0 - 6 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	Samsung Galaxy S23	Snapdragon® 8 Gen 2	QNN	0.547 ms	0 - 6 MB	FP16	NPU	MediaPipe-Face-Detection.so
FaceDetector	Samsung Galaxy S23	Snapdragon® 8 Gen 2	ONNX	0.859 ms	0 - 7 MB	FP16	NPU	MediaPipe-Face-Detection.onnx
FaceDetector	Samsung Galaxy S24	Snapdragon® 8 Gen 3	TFLITE	0.4 ms	0 - 27 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	Samsung Galaxy S24	Snapdragon® 8 Gen 3	QNN	0.397 ms	0 - 29 MB	FP16	NPU	MediaPipe-Face-Detection.so
FaceDetector	Samsung Galaxy S24	Snapdragon® 8 Gen 3	ONNX	0.569 ms	0 - 30 MB	FP16	NPU	MediaPipe-Face-Detection.onnx
FaceDetector	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	TFLITE	0.412 ms	0 - 16 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	QNN	0.409 ms	1 - 21 MB	FP16	NPU	Use Export Script
FaceDetector	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	ONNX	0.588 ms	1 - 18 MB	FP16	NPU	MediaPipe-Face-Detection.onnx
FaceDetector	SA7255P ADP	SA7255P	TFLITE	18.702 ms	0 - 14 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	SA7255P ADP	SA7255P	QNN	18.726 ms	1 - 10 MB	FP16	NPU	Use Export Script
FaceDetector	SA8255 (Proxy)	SA8255P Proxy	TFLITE	0.552 ms	0 - 7 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	SA8255 (Proxy)	SA8255P Proxy	QNN	0.528 ms	1 - 3 MB	FP16	NPU	Use Export Script
FaceDetector	SA8295P ADP	SA8295P	TFLITE	1.17 ms	0 - 19 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	SA8295P ADP	SA8295P	QNN	1.065 ms	1 - 19 MB	FP16	NPU	Use Export Script
FaceDetector	SA8650 (Proxy)	SA8650P Proxy	TFLITE	0.55 ms	0 - 7 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	SA8650 (Proxy)	SA8650P Proxy	QNN	0.532 ms	0 - 2 MB	FP16	NPU	Use Export Script
FaceDetector	SA8775P ADP	SA8775P	TFLITE	1.251 ms	0 - 15 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	SA8775P ADP	SA8775P	QNN	1.179 ms	1 - 11 MB	FP16	NPU	Use Export Script
FaceDetector	QCS8275 (Proxy)	QCS8275 Proxy	TFLITE	18.702 ms	0 - 14 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	QCS8275 (Proxy)	QCS8275 Proxy	QNN	18.726 ms	1 - 10 MB	FP16	NPU	Use Export Script
FaceDetector	QCS8550 (Proxy)	QCS8550 Proxy	TFLITE	0.545 ms	0 - 6 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	QCS8550 (Proxy)	QCS8550 Proxy	QNN	0.529 ms	1 - 3 MB	FP16	NPU	Use Export Script
FaceDetector	QCS9075 (Proxy)	QCS9075 Proxy	TFLITE	1.251 ms	0 - 15 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	QCS9075 (Proxy)	QCS9075 Proxy	QNN	1.179 ms	1 - 11 MB	FP16	NPU	Use Export Script
FaceDetector	QCS8450 (Proxy)	QCS8450 Proxy	TFLITE	0.826 ms	0 - 22 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceDetector	QCS8450 (Proxy)	QCS8450 Proxy	QNN	0.824 ms	1 - 22 MB	FP16	NPU	Use Export Script
FaceDetector	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN	0.668 ms	1 - 1 MB	FP16	NPU	Use Export Script
FaceDetector	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	0.846 ms	2 - 2 MB	FP16	NPU	MediaPipe-Face-Detection.onnx
FaceLandmarkDetector	Samsung Galaxy S23	Snapdragon® 8 Gen 2	TFLITE	0.205 ms	0 - 7 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	Samsung Galaxy S23	Snapdragon® 8 Gen 2	QNN	0.226 ms	0 - 7 MB	FP16	NPU	MediaPipe-Face-Detection.so
FaceLandmarkDetector	Samsung Galaxy S23	Snapdragon® 8 Gen 2	ONNX	0.386 ms	0 - 7 MB	FP16	NPU	MediaPipe-Face-Detection.onnx
FaceLandmarkDetector	Samsung Galaxy S24	Snapdragon® 8 Gen 3	TFLITE	0.151 ms	0 - 21 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	Samsung Galaxy S24	Snapdragon® 8 Gen 3	QNN	0.16 ms	0 - 25 MB	FP16	NPU	MediaPipe-Face-Detection.so
FaceLandmarkDetector	Samsung Galaxy S24	Snapdragon® 8 Gen 3	ONNX	0.277 ms	0 - 23 MB	FP16	NPU	MediaPipe-Face-Detection.onnx
FaceLandmarkDetector	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	TFLITE	0.128 ms	0 - 19 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	QNN	0.168 ms	0 - 14 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	ONNX	0.329 ms	0 - 17 MB	FP16	NPU	MediaPipe-Face-Detection.onnx
FaceLandmarkDetector	SA7255P ADP	SA7255P	TFLITE	3.543 ms	0 - 13 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	SA7255P ADP	SA7255P	QNN	3.557 ms	0 - 10 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	SA8255 (Proxy)	SA8255P Proxy	TFLITE	0.198 ms	0 - 7 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	SA8255 (Proxy)	SA8255P Proxy	QNN	0.22 ms	0 - 3 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	SA8295P ADP	SA8295P	TFLITE	0.578 ms	0 - 17 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	SA8295P ADP	SA8295P	QNN	0.575 ms	0 - 18 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	SA8650 (Proxy)	SA8650P Proxy	TFLITE	0.203 ms	0 - 7 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	SA8650 (Proxy)	SA8650P Proxy	QNN	0.215 ms	0 - 3 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	SA8775P ADP	SA8775P	TFLITE	0.506 ms	0 - 13 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	SA8775P ADP	SA8775P	QNN	0.52 ms	0 - 11 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	QCS8275 (Proxy)	QCS8275 Proxy	TFLITE	3.543 ms	0 - 13 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	QCS8275 (Proxy)	QCS8275 Proxy	QNN	3.557 ms	0 - 10 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	QCS8550 (Proxy)	QCS8550 Proxy	TFLITE	0.206 ms	0 - 4 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	QCS8550 (Proxy)	QCS8550 Proxy	QNN	0.217 ms	0 - 3 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	QCS9075 (Proxy)	QCS9075 Proxy	TFLITE	0.506 ms	0 - 13 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	QCS9075 (Proxy)	QCS9075 Proxy	QNN	0.52 ms	0 - 11 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	QCS8450 (Proxy)	QCS8450 Proxy	TFLITE	0.344 ms	0 - 23 MB	FP16	NPU	MediaPipe-Face-Detection.tflite
FaceLandmarkDetector	QCS8450 (Proxy)	QCS8450 Proxy	QNN	0.38 ms	0 - 22 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN	0.314 ms	0 - 0 MB	FP16	NPU	Use Export Script
FaceLandmarkDetector	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	0.375 ms	1 - 1 MB	FP16	NPU	MediaPipe-Face-Detection.onnx

Installation

Install the package via pip:

pip install qai-hub-models

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.mediapipe_face.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.mediapipe_face.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.mediapipe_face.export

Profiling Results
------------------------------------------------------------
FaceDetector
Device                          : Samsung Galaxy S23 (13)
Runtime                         : TFLITE                 
Estimated inference time (ms)   : 0.6                    
Estimated peak memory usage (MB): [0, 6]                 
Total # Ops                     : 111                    
Compute Unit(s)                 : NPU (111 ops)          

------------------------------------------------------------
FaceLandmarkDetector
Device                          : Samsung Galaxy S23 (13)
Runtime                         : TFLITE                 
Estimated inference time (ms)   : 0.2                    
Estimated peak memory usage (MB): [0, 7]                 
Total # Ops                     : 100                    
Compute Unit(s)                 : NPU (100 ops)

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.mediapipe_face import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S24")

# 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.

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 MediaPipe-Face-Detection's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of MediaPipe-Face-Detection 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.