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: Model_use_case.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 Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
FaceDetector float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 18.759 ms 0 - 14 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN 18.64 ms 1 - 10 MB NPU Use Export Script
FaceDetector float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 0.817 ms 0 - 23 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN 0.808 ms 1 - 26 MB NPU Use Export Script
FaceDetector float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 0.548 ms 0 - 6 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN 0.537 ms 1 - 3 MB NPU Use Export Script
FaceDetector float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 1.081 ms 0 - 17 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN 1.009 ms 1 - 15 MB NPU Use Export Script
FaceDetector float SA7255P ADP Qualcomm® SA7255P TFLITE 18.759 ms 0 - 14 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float SA7255P ADP Qualcomm® SA7255P QNN 18.64 ms 1 - 10 MB NPU Use Export Script
FaceDetector float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 0.553 ms 0 - 6 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN 0.536 ms 1 - 3 MB NPU Use Export Script
FaceDetector float SA8295P ADP Qualcomm® SA8295P TFLITE 1.138 ms 0 - 19 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float SA8295P ADP Qualcomm® SA8295P QNN 1.064 ms 0 - 18 MB NPU Use Export Script
FaceDetector float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 0.549 ms 0 - 7 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN 0.536 ms 1 - 3 MB NPU Use Export Script
FaceDetector float SA8775P ADP Qualcomm® SA8775P TFLITE 1.081 ms 0 - 17 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float SA8775P ADP Qualcomm® SA8775P QNN 1.009 ms 1 - 15 MB NPU Use Export Script
FaceDetector float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 0.554 ms 0 - 6 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN 0.541 ms 2 - 8 MB NPU Use Export Script
FaceDetector float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 0.835 ms 2 - 9 MB NPU MediaPipe-Face-Detection.onnx
FaceDetector float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 0.396 ms 0 - 27 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN 0.39 ms 0 - 27 MB NPU Use Export Script
FaceDetector float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 0.574 ms 0 - 21 MB NPU MediaPipe-Face-Detection.onnx
FaceDetector float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 0.414 ms 0 - 25 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN 0.327 ms 0 - 21 MB NPU Use Export Script
FaceDetector float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 0.495 ms 1 - 24 MB NPU MediaPipe-Face-Detection.onnx
FaceDetector float Snapdragon X Elite CRD Snapdragon® X Elite QNN 0.682 ms 1 - 1 MB NPU Use Export Script
FaceDetector float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 0.891 ms 2 - 2 MB NPU MediaPipe-Face-Detection.onnx
FaceLandmarkDetector float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 0.861 ms 0 - 12 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN 0.877 ms 0 - 10 MB NPU Use Export Script
FaceLandmarkDetector float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 0.338 ms 0 - 24 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN 0.407 ms 0 - 27 MB NPU Use Export Script
FaceLandmarkDetector float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 0.199 ms 0 - 5 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN 0.214 ms 0 - 4 MB NPU Use Export Script
FaceLandmarkDetector float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 0.456 ms 0 - 16 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN 0.447 ms 0 - 15 MB NPU Use Export Script
FaceLandmarkDetector float SA7255P ADP Qualcomm® SA7255P TFLITE 0.861 ms 0 - 12 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float SA7255P ADP Qualcomm® SA7255P QNN 0.877 ms 0 - 10 MB NPU Use Export Script
FaceLandmarkDetector float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 0.194 ms 0 - 8 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN 0.22 ms 0 - 3 MB NPU Use Export Script
FaceLandmarkDetector float SA8295P ADP Qualcomm® SA8295P TFLITE 0.543 ms 0 - 14 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float SA8295P ADP Qualcomm® SA8295P QNN 0.579 ms 0 - 18 MB NPU Use Export Script
FaceLandmarkDetector float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 0.198 ms 0 - 7 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN 0.215 ms 0 - 3 MB NPU Use Export Script
FaceLandmarkDetector float SA8775P ADP Qualcomm® SA8775P TFLITE 0.456 ms 0 - 16 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float SA8775P ADP Qualcomm® SA8775P QNN 0.447 ms 0 - 15 MB NPU Use Export Script
FaceLandmarkDetector float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 0.197 ms 0 - 8 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN 0.223 ms 0 - 8 MB NPU Use Export Script
FaceLandmarkDetector float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 0.369 ms 0 - 7 MB NPU MediaPipe-Face-Detection.onnx
FaceLandmarkDetector float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 0.142 ms 0 - 19 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN 0.165 ms 0 - 25 MB NPU Use Export Script
FaceLandmarkDetector float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 0.287 ms 0 - 27 MB NPU MediaPipe-Face-Detection.onnx
FaceLandmarkDetector float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 0.15 ms 0 - 19 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN 0.16 ms 0 - 17 MB NPU Use Export Script
FaceLandmarkDetector float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 0.248 ms 0 - 19 MB NPU MediaPipe-Face-Detection.onnx
FaceLandmarkDetector float Snapdragon X Elite CRD Snapdragon® X Elite QNN 0.297 ms 0 - 0 MB NPU Use Export Script
FaceLandmarkDetector float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 0.378 ms 0 - 0 MB NPU MediaPipe-Face-Detection.onnx
FaceDetector w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 0.667 ms 0 - 17 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN 0.665 ms 0 - 10 MB NPU Use Export Script
FaceDetector w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 0.36 ms 0 - 26 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN 0.35 ms 0 - 26 MB NPU Use Export Script
FaceDetector w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 0.224 ms 0 - 12 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN 0.233 ms 0 - 3 MB NPU Use Export Script
FaceDetector w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 0.449 ms 0 - 19 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN 0.423 ms 0 - 15 MB NPU Use Export Script
FaceDetector w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) TFLITE 0.608 ms 0 - 26 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN 0.654 ms 0 - 12 MB NPU Use Export Script
FaceDetector w8a8 RB5 (Proxy) Qualcomm® QCS8250 (Proxy) TFLITE 5.175 ms 0 - 3 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 SA7255P ADP Qualcomm® SA7255P TFLITE 0.667 ms 0 - 17 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 SA7255P ADP Qualcomm® SA7255P QNN 0.665 ms 0 - 10 MB NPU Use Export Script
FaceDetector w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 0.234 ms 0 - 12 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN 0.239 ms 0 - 3 MB NPU Use Export Script
FaceDetector w8a8 SA8295P ADP Qualcomm® SA8295P TFLITE 0.59 ms 0 - 22 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 SA8295P ADP Qualcomm® SA8295P QNN 0.556 ms 0 - 20 MB NPU Use Export Script
FaceDetector w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 0.237 ms 0 - 12 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN 0.224 ms 0 - 3 MB NPU Use Export Script
FaceDetector w8a8 SA8775P ADP Qualcomm® SA8775P TFLITE 0.449 ms 0 - 19 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 SA8775P ADP Qualcomm® SA8775P QNN 0.423 ms 0 - 15 MB NPU Use Export Script
FaceDetector w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 0.237 ms 0 - 13 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN 0.227 ms 0 - 12 MB NPU Use Export Script
FaceDetector w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 0.381 ms 0 - 12 MB NPU MediaPipe-Face-Detection.onnx
FaceDetector w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 0.155 ms 0 - 31 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN 0.145 ms 0 - 28 MB NPU Use Export Script
FaceDetector w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 0.241 ms 0 - 35 MB NPU MediaPipe-Face-Detection.onnx
FaceDetector w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 0.189 ms 0 - 23 MB NPU MediaPipe-Face-Detection.tflite
FaceDetector w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN 0.132 ms 0 - 23 MB NPU Use Export Script
FaceDetector w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 0.283 ms 0 - 26 MB NPU MediaPipe-Face-Detection.onnx
FaceDetector w8a8 Snapdragon X Elite CRD Snapdragon® X Elite QNN 0.324 ms 0 - 0 MB NPU Use Export Script
FaceDetector w8a8 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 0.374 ms 1 - 1 MB NPU MediaPipe-Face-Detection.onnx
FaceLandmarkDetector w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 0.808 ms 0 - 13 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN 0.479 ms 0 - 10 MB NPU Use Export Script
FaceLandmarkDetector w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 0.261 ms 0 - 22 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN 0.269 ms 0 - 22 MB NPU Use Export Script
FaceLandmarkDetector w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 0.159 ms 0 - 10 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN 0.152 ms 0 - 3 MB NPU Use Export Script
FaceLandmarkDetector w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 0.346 ms 0 - 15 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN 0.332 ms 0 - 15 MB NPU Use Export Script
FaceLandmarkDetector w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) TFLITE 0.374 ms 0 - 17 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN 0.424 ms 0 - 12 MB NPU Use Export Script
FaceLandmarkDetector w8a8 RB5 (Proxy) Qualcomm® QCS8250 (Proxy) TFLITE 2.559 ms 0 - 3 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 SA7255P ADP Qualcomm® SA7255P TFLITE 0.808 ms 0 - 13 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 SA7255P ADP Qualcomm® SA7255P QNN 0.479 ms 0 - 10 MB NPU Use Export Script
FaceLandmarkDetector w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 0.161 ms 0 - 11 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN 0.151 ms 0 - 2 MB NPU Use Export Script
FaceLandmarkDetector w8a8 SA8295P ADP Qualcomm® SA8295P TFLITE 0.455 ms 0 - 18 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 SA8295P ADP Qualcomm® SA8295P QNN 0.461 ms 0 - 18 MB NPU Use Export Script
FaceLandmarkDetector w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 0.156 ms 0 - 11 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN 0.144 ms 0 - 3 MB NPU Use Export Script
FaceLandmarkDetector w8a8 SA8775P ADP Qualcomm® SA8775P TFLITE 0.346 ms 0 - 15 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 SA8775P ADP Qualcomm® SA8775P QNN 0.332 ms 0 - 15 MB NPU Use Export Script
FaceLandmarkDetector w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 0.164 ms 0 - 5 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN 0.149 ms 0 - 10 MB NPU Use Export Script
FaceLandmarkDetector w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 0.303 ms 0 - 12 MB NPU MediaPipe-Face-Detection.onnx
FaceLandmarkDetector w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 0.121 ms 0 - 23 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN 0.12 ms 0 - 21 MB NPU Use Export Script
FaceLandmarkDetector w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 0.213 ms 0 - 24 MB NPU MediaPipe-Face-Detection.onnx
FaceLandmarkDetector w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 0.124 ms 0 - 18 MB NPU MediaPipe-Face-Detection.tflite
FaceLandmarkDetector w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN 0.119 ms 0 - 19 MB NPU Use Export Script
FaceLandmarkDetector w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 0.248 ms 0 - 21 MB NPU MediaPipe-Face-Detection.onnx
FaceLandmarkDetector w8a8 Snapdragon X Elite CRD Snapdragon® X Elite QNN 0.241 ms 0 - 0 MB NPU Use Export Script
FaceLandmarkDetector w8a8 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 0.301 ms 0 - 0 MB 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                          : cs_8275 (ANDROID 14)                 
Runtime                         : TFLITE                               
Estimated inference time (ms)   : 0.7                                  
Estimated peak memory usage (MB): [0, 17]                              
Total # Ops                     : 114                                  
Compute Unit(s)                 : npu (114 ops) gpu (0 ops) cpu (0 ops)

------------------------------------------------------------
FaceLandmarkDetector
Device                          : cs_8275 (ANDROID 14)                 
Runtime                         : TFLITE                               
Estimated inference time (ms)   : 0.8                                  
Estimated peak memory usage (MB): [0, 13]                              
Total # Ops                     : 108                                  
Compute Unit(s)                 : npu (108 ops) gpu (0 ops) cpu (0 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

Downloads last month
564
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support