MediaPipe-Pose-Estimation: Optimized for Mobile Deployment
Detect and track human face, hand, and torso in real-time images and video streams
The MediaPipe Pose Landmark Detector is a machine learning pipeline that predicts bounding boxes and pose skeletons of the face, hands, and torso in an image.
This model is an implementation of MediaPipe-Pose-Estimation found here.
This repository provides scripts to run MediaPipe-Pose-Estimation 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: 256x256
- Number of parameters (PoseDetector): 815K
- Model size (PoseDetector) (float): 3.14 MB
- Number of parameters (PoseLandmarkDetector): 3.36M
- Model size (PoseLandmarkDetector) (float): 12.9 MB
| Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |
|---|---|---|---|---|---|---|---|---|
| PoseDetector | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 5.795 ms | 0 - 123 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 5.751 ms | 0 - 122 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 2.184 ms | 0 - 144 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 2.191 ms | 0 - 146 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 0.856 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 0.855 ms | 0 - 3 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 1.133 ms | 0 - 5 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 1.64 ms | 0 - 122 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 1.63 ms | 0 - 123 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 5.795 ms | 0 - 123 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 5.751 ms | 0 - 122 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 0.849 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 0.851 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 2.572 ms | 0 - 126 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 2.536 ms | 0 - 126 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 0.85 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 0.854 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 1.64 ms | 0 - 122 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 1.63 ms | 0 - 123 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.632 ms | 0 - 145 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.626 ms | 0 - 147 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 0.781 ms | 0 - 122 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.514 ms | 0 - 127 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 0.514 ms | 0 - 127 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 0.648 ms | 0 - 102 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 0.401 ms | 0 - 127 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 0.403 ms | 0 - 126 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 0.593 ms | 0 - 102 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 1.011 ms | 0 - 0 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 1.048 ms | 2 - 2 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 3.161 ms | 0 - 136 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 3.165 ms | 1 - 130 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 1.178 ms | 0 - 175 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 1.179 ms | 1 - 170 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 0.799 ms | 0 - 3 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 0.814 ms | 1 - 3 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 1.142 ms | 0 - 9 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 4.758 ms | 0 - 136 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 4.749 ms | 1 - 131 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 3.161 ms | 0 - 136 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 3.165 ms | 1 - 130 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 0.795 ms | 0 - 3 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 0.812 ms | 1 - 3 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 1.416 ms | 0 - 140 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 1.403 ms | 0 - 133 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 0.821 ms | 0 - 3 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 0.807 ms | 1 - 3 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 4.758 ms | 0 - 136 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 4.749 ms | 1 - 131 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.577 ms | 0 - 177 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.583 ms | 1 - 174 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 0.794 ms | 0 - 151 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.495 ms | 0 - 142 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 0.495 ms | 1 - 133 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 0.708 ms | 0 - 112 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 0.426 ms | 0 - 140 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 0.423 ms | 1 - 137 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 0.641 ms | 0 - 113 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 1.078 ms | 1 - 1 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 1.168 ms | 7 - 7 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® Qcm6690 | TFLITE | 3.207 ms | 0 - 125 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® Qcm6690 | QNN_DLC | 3.203 ms | 0 - 126 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® Qcm6690 | ONNX | 10.247 ms | 23 - 39 MB | CPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | TFLITE | 1.512 ms | 0 - 5 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | QNN_DLC | 1.542 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | ONNX | 13.892 ms | 11 - 15 MB | CPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 1.029 ms | 0 - 121 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 1.004 ms | 0 - 122 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 0.587 ms | 0 - 139 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 0.57 ms | 0 - 144 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 0.316 ms | 0 - 3 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 0.32 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 0.534 ms | 0 - 5 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 0.535 ms | 0 - 121 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 0.505 ms | 0 - 123 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 7.1 ms | 0 - 9 MB | GPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 8.701 ms | 8 - 13 MB | CPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 1.029 ms | 0 - 121 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 1.004 ms | 0 - 122 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 0.311 ms | 0 - 3 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 0.318 ms | 0 - 3 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 0.95 ms | 0 - 128 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 0.923 ms | 0 - 128 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 0.321 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 0.316 ms | 0 - 3 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 0.535 ms | 0 - 121 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 0.505 ms | 0 - 123 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.229 ms | 0 - 141 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.226 ms | 0 - 138 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 0.355 ms | 0 - 118 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.189 ms | 0 - 122 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 0.19 ms | 0 - 126 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 0.29 ms | 0 - 105 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | TFLITE | 0.397 ms | 0 - 125 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | QNN_DLC | 0.396 ms | 0 - 127 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | ONNX | 9.47 ms | 9 - 25 MB | CPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 0.17 ms | 0 - 124 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseDetector | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 0.166 ms | 0 - 125 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 0.284 ms | 0 - 104 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseDetector | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 0.428 ms | 0 - 0 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseDetector | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 0.457 ms | 0 - 0 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® Qcm6690 | TFLITE | 1.635 ms | 0 - 133 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® Qcm6690 | QNN_DLC | 1.603 ms | 0 - 134 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® Qcm6690 | ONNX | 9.281 ms | 9 - 27 MB | CPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | TFLITE | 1.119 ms | 0 - 9 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | QNN_DLC | 1.068 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | ONNX | 19.04 ms | 8 - 15 MB | CPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 0.83 ms | 0 - 130 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 0.795 ms | 0 - 131 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 0.455 ms | 0 - 156 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 0.477 ms | 0 - 159 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 0.31 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 0.309 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 0.55 ms | 0 - 7 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 0.543 ms | 0 - 130 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 0.516 ms | 0 - 132 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 5.82 ms | 0 - 23 MB | GPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 7.752 ms | 6 - 13 MB | CPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 0.83 ms | 0 - 130 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 0.795 ms | 0 - 131 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 0.308 ms | 0 - 3 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 0.307 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 0.723 ms | 0 - 138 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 0.718 ms | 0 - 138 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 0.31 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 0.305 ms | 0 - 2 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 0.543 ms | 0 - 130 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 0.516 ms | 0 - 132 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.212 ms | 0 - 154 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.211 ms | 0 - 156 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 0.365 ms | 0 - 138 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.194 ms | 0 - 134 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 0.19 ms | 0 - 136 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 0.329 ms | 0 - 120 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | TFLITE | 0.322 ms | 0 - 134 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | QNN_DLC | 0.317 ms | 0 - 137 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | ONNX | 8.514 ms | 9 - 28 MB | CPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 0.158 ms | 0 - 134 MB | NPU | MediaPipe-Pose-Estimation.tflite |
| PoseLandmarkDetector | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 0.16 ms | 0 - 137 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 0.306 ms | 0 - 120 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
| PoseLandmarkDetector | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 0.44 ms | 0 - 0 MB | NPU | MediaPipe-Pose-Estimation.dlc |
| PoseLandmarkDetector | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 0.457 ms | 4 - 4 MB | NPU | MediaPipe-Pose-Estimation.onnx.zip |
Installation
Install the package via pip:
pip install qai-hub-models
Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device
Sign-in to Qualcomm® AI Hub Workbench 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_pose.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_pose.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_pose.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.mediapipe_pose 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 Workbench. Sign up for access.
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 MediaPipe-Pose-Estimation's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of MediaPipe-Pose-Estimation 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.
- Downloads last month
- 540
