Facial-Landmark-Detection: Optimized for Mobile Deployment
Real-time 3D facial landmark detection optimized for mobile and edge
Detects facial landmarks (eg, nose, mouth, etc.). This model's architecture was developed by Qualcomm. The model was trained by Qualcomm on a proprietary dataset of faces, but can be used on any image.
This repository provides scripts to run Facial-Landmark-Detection 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:
- Input resolution: 128x128
- Number of parameters: 5.42M
- Model size (float): 20.7 MB
- Model size (w8a8): 5.27 MB
| Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |
|---|---|---|---|---|---|---|---|---|
| Facial-Landmark-Detection | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 1.147 ms | 0 - 17 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 1.141 ms | 0 - 15 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 0.397 ms | 0 - 37 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 0.548 ms | 0 - 23 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 0.29 ms | 0 - 104 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 0.292 ms | 0 - 46 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 0.497 ms | 0 - 48 MB | NPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 0.514 ms | 0 - 17 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 1.73 ms | 0 - 16 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 1.147 ms | 0 - 17 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 1.141 ms | 0 - 15 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 0.289 ms | 0 - 100 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 0.294 ms | 0 - 49 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 0.664 ms | 0 - 22 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 0.638 ms | 0 - 21 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 0.288 ms | 0 - 100 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 0.383 ms | 0 - 36 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 0.514 ms | 0 - 17 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 1.73 ms | 0 - 16 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.215 ms | 0 - 41 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.216 ms | 0 - 25 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 0.34 ms | 0 - 24 MB | NPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.197 ms | 0 - 21 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 0.189 ms | 0 - 19 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 0.319 ms | 0 - 18 MB | NPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 0.202 ms | 0 - 24 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 0.183 ms | 0 - 18 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 0.32 ms | 0 - 21 MB | NPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 0.356 ms | 35 - 35 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 0.391 ms | 11 - 11 MB | NPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 0.473 ms | 0 - 16 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 0.433 ms | 0 - 16 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 0.222 ms | 0 - 32 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 0.236 ms | 0 - 36 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 0.178 ms | 0 - 42 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 0.169 ms | 0 - 43 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 0.369 ms | 0 - 15 MB | NPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 0.331 ms | 0 - 15 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 0.309 ms | 0 - 16 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 0.546 ms | 0 - 26 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | ONNX | 1.839 ms | 0 - 12 MB | CPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 2.037 ms | 0 - 3 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 1.525 ms | 0 - 20 MB | CPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 0.473 ms | 0 - 16 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 0.433 ms | 0 - 16 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 0.174 ms | 0 - 42 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 0.173 ms | 0 - 32 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 0.455 ms | 0 - 23 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 0.46 ms | 0 - 21 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 0.177 ms | 0 - 43 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 0.165 ms | 0 - 43 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 0.331 ms | 0 - 15 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 0.309 ms | 0 - 16 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.138 ms | 0 - 37 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.138 ms | 0 - 35 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 0.246 ms | 0 - 35 MB | NPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.121 ms | 0 - 18 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 0.109 ms | 0 - 20 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 0.21 ms | 0 - 21 MB | NPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 0.117 ms | 0 - 18 MB | NPU | Facial-Landmark-Detection.tflite |
| Facial-Landmark-Detection | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 0.119 ms | 0 - 18 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 0.223 ms | 0 - 20 MB | NPU | Facial-Landmark-Detection.onnx.zip |
| Facial-Landmark-Detection | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 0.23 ms | 31 - 31 MB | NPU | Facial-Landmark-Detection.dlc |
| Facial-Landmark-Detection | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 0.245 ms | 5 - 5 MB | NPU | Facial-Landmark-Detection.onnx.zip |
Installation
Install the package via pip:
pip install "qai-hub-models[facemap-3dmm]"
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.facemap_3dmm.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.facemap_3dmm.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.facemap_3dmm.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.facemap_3dmm 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.facemap_3dmm.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.facemap_3dmm.demo -- --eval-mode on-device
Deploying compiled model to Android
The models can be deployed using multiple runtimes:
TensorFlow Lite (
.tfliteexport): This tutorial provides a guide to deploy the .tflite model in an Android application.QNN (
.soexport ): This sample app provides instructions on how to use the.soshared library in an Android application.
View on Qualcomm® AI Hub
Get more details on Facial-Landmark-Detection's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of Facial-Landmark-Detection can be found here.
- The license for the compiled assets for on-device deployment can be found here
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.
- Downloads last month
- 784
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
🙋
Ask for provider support