--- library_name: pytorch license: apache-2.0 tags: - real_time - quantized - android pipeline_tag: object-detection --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mediapipe_face_quantized/web-assets/model_demo.png) # MediaPipe-Face-Detection-Quantized: 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-Quantized found [here](https://github.com/zmurez/MediaPipePyTorch/). This repository provides scripts to run MediaPipe-Face-Detection-Quantized on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/mediapipe_face_quantized). ### Model Details - **Model Type:** Object detection - **Model Stats:** - Input resolution: 256x256 - Number of output classes: 6 - Number of parameters (MediaPipeFaceDetector): 135K - Model size (MediaPipeFaceDetector): 255 KB - Number of parameters (MediaPipeFaceLandmarkDetector): 603K - Model size (MediaPipeFaceLandmarkDetector): 746 KB | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | FaceDetectorQuantizable | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 0.262 ms | 0 - 11 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 0.243 ms | 0 - 12 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.so](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.so) | | FaceDetectorQuantizable | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 0.48 ms | 2 - 13 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.onnx](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.onnx) | | FaceDetectorQuantizable | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 0.18 ms | 0 - 24 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 0.165 ms | 0 - 28 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.so](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.so) | | FaceDetectorQuantizable | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 0.309 ms | 0 - 24 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.onnx](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.onnx) | | FaceDetectorQuantizable | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 0.157 ms | 0 - 20 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 0.131 ms | 0 - 20 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 0.282 ms | 0 - 20 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.onnx](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.onnx) | | FaceDetectorQuantizable | SA7255P ADP | SA7255P | TFLITE | 2.023 ms | 0 - 17 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | SA7255P ADP | SA7255P | QNN | 1.856 ms | 0 - 10 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 0.27 ms | 0 - 11 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | SA8255 (Proxy) | SA8255P Proxy | QNN | 0.241 ms | 0 - 3 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | SA8295P ADP | SA8295P | TFLITE | 0.669 ms | 0 - 21 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | SA8295P ADP | SA8295P | QNN | 0.576 ms | 0 - 18 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 0.264 ms | 0 - 11 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | SA8650 (Proxy) | SA8650P Proxy | QNN | 0.239 ms | 0 - 2 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | SA8775P ADP | SA8775P | TFLITE | 0.55 ms | 0 - 17 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | SA8775P ADP | SA8775P | QNN | 0.51 ms | 0 - 10 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | RB3 Gen 2 (Proxy) | QCS6490 Proxy | TFLITE | 0.697 ms | 0 - 22 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | RB3 Gen 2 (Proxy) | QCS6490 Proxy | QNN | 0.64 ms | 0 - 13 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | RB5 (Proxy) | QCS8250 Proxy | TFLITE | 4.791 ms | 0 - 2 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | QCS8275 (Proxy) | QCS8275 Proxy | TFLITE | 2.023 ms | 0 - 17 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | QCS8275 (Proxy) | QCS8275 Proxy | QNN | 1.856 ms | 0 - 10 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 0.266 ms | 0 - 11 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 0.237 ms | 0 - 3 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | QCS9075 (Proxy) | QCS9075 Proxy | TFLITE | 0.55 ms | 0 - 17 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | QCS9075 (Proxy) | QCS9075 Proxy | QNN | 0.51 ms | 0 - 10 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 0.374 ms | 0 - 28 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.tflite) | | FaceDetectorQuantizable | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 0.362 ms | 0 - 25 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 0.323 ms | 0 - 0 MB | INT8 | NPU | Use Export Script | | FaceDetectorQuantizable | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 0.418 ms | 1 - 1 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.onnx](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceDetectorQuantizable.onnx) | | FaceLandmarkDetectorQuantizable | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 0.182 ms | 0 - 11 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 0.153 ms | 0 - 11 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.so](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.so) | | FaceLandmarkDetectorQuantizable | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 0.47 ms | 0 - 9 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.onnx](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.onnx) | | FaceLandmarkDetectorQuantizable | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 0.139 ms | 0 - 22 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 0.118 ms | 0 - 19 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.so](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.so) | | FaceLandmarkDetectorQuantizable | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 0.325 ms | 0 - 26 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.onnx](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.onnx) | | FaceLandmarkDetectorQuantizable | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 0.134 ms | 0 - 19 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 0.115 ms | 0 - 19 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 0.347 ms | 0 - 17 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.onnx](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.onnx) | | FaceLandmarkDetectorQuantizable | SA7255P ADP | SA7255P | TFLITE | 0.869 ms | 0 - 12 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | SA7255P ADP | SA7255P | QNN | 0.798 ms | 0 - 10 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 0.181 ms | 0 - 5 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | SA8255 (Proxy) | SA8255P Proxy | QNN | 0.159 ms | 0 - 2 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | SA8295P ADP | SA8295P | TFLITE | 0.502 ms | 0 - 17 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | SA8295P ADP | SA8295P | QNN | 0.464 ms | 0 - 18 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 0.185 ms | 0 - 10 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | SA8650 (Proxy) | SA8650P Proxy | QNN | 0.15 ms | 0 - 2 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | SA8775P ADP | SA8775P | TFLITE | 0.429 ms | 0 - 12 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | SA8775P ADP | SA8775P | QNN | 0.391 ms | 0 - 10 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | RB3 Gen 2 (Proxy) | QCS6490 Proxy | TFLITE | 0.403 ms | 0 - 21 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | RB3 Gen 2 (Proxy) | QCS6490 Proxy | QNN | 0.392 ms | 0 - 15 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | RB5 (Proxy) | QCS8250 Proxy | TFLITE | 2.793 ms | 0 - 3 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | QCS8275 (Proxy) | QCS8275 Proxy | TFLITE | 0.869 ms | 0 - 12 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | QCS8275 (Proxy) | QCS8275 Proxy | QNN | 0.798 ms | 0 - 10 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 0.174 ms | 0 - 10 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 0.154 ms | 0 - 2 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | QCS9075 (Proxy) | QCS9075 Proxy | TFLITE | 0.429 ms | 0 - 12 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | QCS9075 (Proxy) | QCS9075 Proxy | QNN | 0.391 ms | 0 - 10 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 0.281 ms | 0 - 23 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.tflite](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.tflite) | | FaceLandmarkDetectorQuantizable | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 0.256 ms | 0 - 24 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 0.228 ms | 1 - 1 MB | INT8 | NPU | Use Export Script | | FaceLandmarkDetectorQuantizable | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 0.479 ms | 0 - 0 MB | INT8 | NPU | [MediaPipe-Face-Detection-Quantized.onnx](https://huggingface.co/qualcomm/MediaPipe-Face-Detection-Quantized/blob/main/FaceLandmarkDetectorQuantizable.onnx) | ## Installation Install the package via pip: ```bash pip install "qai-hub-models[mediapipe-face-quantized]" ``` ## Configure Qualcomm® AI Hub to run this model on a cloud-hosted device Sign-in to [Qualcomm® AI Hub](https://app.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://app.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.mediapipe_face_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.mediapipe_face_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. ```bash python -m qai_hub_models.models.mediapipe_face_quantized.export ``` ``` Profiling Results ------------------------------------------------------------ FaceDetectorQuantizable Device : Samsung Galaxy S23 (13) Runtime : TFLITE Estimated inference time (ms) : 0.3 Estimated peak memory usage (MB): [0, 11] Total # Ops : 121 Compute Unit(s) : NPU (121 ops) ------------------------------------------------------------ FaceLandmarkDetectorQuantizable Device : Samsung Galaxy S23 (13) Runtime : TFLITE Estimated inference time (ms) : 0.2 Estimated peak memory usage (MB): [0, 11] Total # Ops : 116 Compute Unit(s) : NPU (116 ops) ``` ## 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 MediaPipe-Face-Detection-Quantized's performance across various devices [here](https://aihub.qualcomm.com/models/mediapipe_face_quantized). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of MediaPipe-Face-Detection-Quantized can be found [here](https://github.com/zmurez/MediaPipePyTorch/blob/master/LICENSE). * 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 * [BlazeFace: Sub-millisecond Neural Face Detection on Mobile GPUs](https://arxiv.org/abs/1907.05047) * [Source Model Implementation](https://github.com/zmurez/MediaPipePyTorch/) ## 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).