--- library_name: pytorch license: other tags: - real_time - android pipeline_tag: object-detection --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/hrnet_face/web-assets/model_demo.png) # HRNetFace: Optimized for Mobile Deployment ## Comprehensive facial analysis by extracting face features Detects attributes (liveness, eye closeness, mask presence, glasses presence, sunglasses presence) that apply to a given face. This model is an implementation of HRNetFace found [here](https://github.com/HRNet/HRNet-Facial-Landmark-Detection). This repository provides scripts to run HRNetFace on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/hrnet_face). ### Model Details - **Model Type:** Model_use_case.object_detection - **Model Stats:** - Model checkpoint: HR18-COFW.pth - Input resolution: 256x256 - Number of parameters: 9.68M - Model size (float): 36.87MB - Model size (w8a8): 17.7 MB | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | HRNetFace | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 15.597 ms | 0 - 60 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 15.505 ms | 1 - 46 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 4.483 ms | 0 - 63 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 4.846 ms | 1 - 48 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 3.141 ms | 0 - 132 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 3.233 ms | 0 - 13 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 3.322 ms | 0 - 45 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.onnx.zip) | | HRNetFace | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 21.582 ms | 0 - 60 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 5.119 ms | 1 - 48 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 15.597 ms | 0 - 60 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 15.505 ms | 1 - 46 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 3.151 ms | 0 - 131 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 3.241 ms | 1 - 14 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 5.53 ms | 0 - 54 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 5.587 ms | 1 - 45 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 3.146 ms | 0 - 129 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 3.247 ms | 1 - 14 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 21.582 ms | 0 - 60 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 5.119 ms | 1 - 48 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 2.219 ms | 0 - 66 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 2.336 ms | 0 - 51 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 2.324 ms | 0 - 68 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.onnx.zip) | | HRNetFace | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 1.748 ms | 0 - 63 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 1.776 ms | 1 - 52 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 1.85 ms | 0 - 56 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.onnx.zip) | | HRNetFace | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 1.417 ms | 0 - 62 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.tflite) | | HRNetFace | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 1.408 ms | 0 - 52 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 1.558 ms | 1 - 58 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.onnx.zip) | | HRNetFace | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 3.628 ms | 27 - 27 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.dlc) | | HRNetFace | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 3.257 ms | 30 - 30 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace.onnx.zip) | | HRNetFace | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | TFLITE | 3.535 ms | 0 - 18 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | QNN_DLC | 3.72 ms | 0 - 108 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | ONNX | 91.59 ms | 18 - 35 MB | CPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) | | HRNetFace | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 3.265 ms | 0 - 46 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 3.38 ms | 0 - 46 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 1.456 ms | 0 - 60 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 1.907 ms | 0 - 56 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 1.295 ms | 0 - 47 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 1.38 ms | 0 - 17 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 1.647 ms | 0 - 21 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) | | HRNetFace | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 1.721 ms | 0 - 46 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 6.892 ms | 0 - 45 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 19.484 ms | 0 - 3 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 46.443 ms | 16 - 31 MB | CPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) | | HRNetFace | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 3.265 ms | 0 - 46 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 3.38 ms | 0 - 46 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 1.298 ms | 0 - 46 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 1.369 ms | 0 - 18 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 2.139 ms | 0 - 53 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 2.272 ms | 0 - 53 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 1.301 ms | 0 - 45 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 1.376 ms | 0 - 18 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 1.721 ms | 0 - 46 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 6.892 ms | 0 - 45 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.884 ms | 0 - 65 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.952 ms | 0 - 59 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 1.071 ms | 0 - 74 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) | | HRNetFace | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.715 ms | 0 - 46 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 0.67 ms | 0 - 52 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 0.886 ms | 0 - 58 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) | | HRNetFace | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | TFLITE | 1.533 ms | 0 - 55 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | QNN_DLC | 1.66 ms | 0 - 57 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | ONNX | 45.713 ms | 20 - 45 MB | CPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) | | HRNetFace | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 0.605 ms | 0 - 53 MB | NPU | [HRNetFace.tflite](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.tflite) | | HRNetFace | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 0.55 ms | 0 - 52 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 0.792 ms | 0 - 60 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) | | HRNetFace | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 1.615 ms | 36 - 36 MB | NPU | [HRNetFace.dlc](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.dlc) | | HRNetFace | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 1.545 ms | 15 - 15 MB | NPU | [HRNetFace.onnx.zip](https://huggingface.co/qualcomm/HRNetFace/blob/main/HRNetFace_w8a8.onnx.zip) | ## Installation Install the package via pip: ```bash # NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported. pip install "qai-hub-models[hrnet-face]" ``` ## Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device Sign-in to [Qualcomm® AI Hub Workbench](https://workbench.aihub.qualcomm.com/) 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. ```bash qai-hub configure --api_token API_TOKEN ``` Navigate to [docs](https://workbench.aihub.qualcomm.com/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. ```bash python -m qai_hub_models.models.hrnet_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.hrnet_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. ```bash python -m qai_hub_models.models.hrnet_face.export ``` ## How does this work? This [export script](https://aihub.qualcomm.com/models/hrnet_face/qai_hub_models/models/HRNetFace/export.py) leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) 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. ```python import torch import qai_hub as hub from qai_hub_models.models.hrnet_face 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. ```python 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. ```python 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 Workbench. [Sign up for access](https://myaccount.qualcomm.com/signup). ## Run demo on a cloud-hosted device You can also run the demo on-device. ```bash python -m qai_hub_models.models.hrnet_face.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_face.demo -- --eval-mode on-device ``` ## Deploying compiled model to Android The models can be deployed using multiple runtimes: - TensorFlow Lite (`.tflite` export): [This tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a guide to deploy the .tflite model in an Android application. - QNN (`.so` export ): This [sample app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html) provides instructions on how to use the `.so` shared library in an Android application. ## View on Qualcomm® AI Hub Get more details on HRNetFace's performance across various devices [here](https://aihub.qualcomm.com/models/hrnet_face). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of HRNetFace can be found [here](https://github.com/HRNet/HRNet-Facial-Landmark-Detection/blob/master/LICENCE). * The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf) ## References * [Deep High-Resolution Representation Learning for Visual Recognition](https://arxiv.org/abs/1908.07919) * [Source Model Implementation](https://github.com/HRNet/HRNet-Facial-Landmark-Detection) ## Community * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI. * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).