MediaPipe-Selfie-Segmentation: Optimized for Qualcomm Devices
Light-weight model that segments a person from the background in square or landscape selfie and video conference imagery.
This is based on the implementation of MediaPipe-Selfie-Segmentation found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.
Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.
Getting Started
There are two ways to deploy this model on your device:
Option 1: Download Pre-Exported Models
Below are pre-exported model assets ready for deployment.
| Runtime | Precision | Chipset | SDK Versions | Download |
|---|---|---|---|---|
| ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| QNN_DLC | float | Universal | QAIRT 2.43 | Download |
| QNN_DLC | w8a16 | Universal | QAIRT 2.43 | Download |
| QNN_DLC | w8a8 | Universal | QAIRT 2.43 | Download |
| TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | Download |
| TFLITE | w8a8 | Universal | QAIRT 2.43, TFLite 2.17.0 | Download |
For more device-specific assets and performance metrics, visit MediaPipe-Selfie-Segmentation on Qualcomm® AI Hub.
Option 2: Export with Custom Configurations
Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:
- Custom weights (e.g., fine-tuned checkpoints)
- Custom input shapes
- Target device and runtime configurations
This option is ideal if you need to customize the model beyond the default configuration provided here.
See our repository for MediaPipe-Selfie-Segmentation on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.semantic_segmentation
Model Stats:
- Model checkpoint: Square
- Input resolution (Square): 256x256
- Input resolution (Landscape): 144x256
- Number of output classes: 6
- Number of parameters: 106K
- Model size (float): 447 KB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| MediaPipe-Selfie-Segmentation | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.37 ms | 0 - 31 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | float | Snapdragon® X2 Elite | 0.383 ms | 0 - 0 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | float | Snapdragon® X Elite | 0.923 ms | 2 - 2 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.521 ms | 0 - 37 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | float | Qualcomm® QCS8550 (Proxy) | 0.779 ms | 1 - 2 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | float | Qualcomm® QCS9075 | 0.984 ms | 1 - 4 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.414 ms | 0 - 24 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.386 ms | 0 - 37 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a16 | Snapdragon® X2 Elite | 0.389 ms | 0 - 0 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a16 | Snapdragon® X Elite | 0.981 ms | 1 - 1 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.556 ms | 0 - 41 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a16 | Qualcomm® QCS6490 | 31.036 ms | 14 - 17 MB | CPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 0.838 ms | 0 - 3 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a16 | Qualcomm® QCS9075 | 0.967 ms | 0 - 3 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a16 | Qualcomm® QCM6690 | 11.785 ms | 15 - 22 MB | CPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.449 ms | 0 - 30 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 8.023 ms | 14 - 22 MB | CPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.192 ms | 0 - 30 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a8 | Snapdragon® X2 Elite | 0.191 ms | 0 - 0 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a8 | Snapdragon® X Elite | 0.477 ms | 1 - 1 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.253 ms | 0 - 36 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a8 | Qualcomm® QCS6490 | 10.053 ms | 7 - 11 MB | CPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.361 ms | 0 - 5 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a8 | Qualcomm® QCS9075 | 0.516 ms | 0 - 3 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a8 | Qualcomm® QCM6690 | 5.955 ms | 5 - 13 MB | CPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.205 ms | 0 - 25 MB | NPU |
| MediaPipe-Selfie-Segmentation | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 4.14 ms | 0 - 8 MB | CPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.316 ms | 1 - 30 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Snapdragon® X2 Elite | 0.487 ms | 1 - 1 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Snapdragon® X Elite | 0.822 ms | 1 - 1 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.457 ms | 0 - 37 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 1.769 ms | 1 - 26 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 0.694 ms | 1 - 2 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Qualcomm® SA8775P | 3.873 ms | 0 - 26 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Qualcomm® QCS9075 | 0.843 ms | 1 - 3 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 0.927 ms | 0 - 39 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Qualcomm® SA7255P | 1.769 ms | 1 - 26 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Qualcomm® SA8295P | 1.286 ms | 0 - 25 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.36 ms | 1 - 26 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.269 ms | 0 - 32 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 0.38 ms | 0 - 0 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Snapdragon® X Elite | 0.755 ms | 0 - 0 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.428 ms | 0 - 36 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Qualcomm® QCS6490 | 1.738 ms | 0 - 3 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 1.352 ms | 0 - 26 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 0.605 ms | 0 - 12 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Qualcomm® SA8775P | 0.838 ms | 0 - 29 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 0.723 ms | 0 - 2 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 2.08 ms | 0 - 138 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 0.758 ms | 0 - 40 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Qualcomm® SA7255P | 1.352 ms | 0 - 26 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Qualcomm® SA8295P | 1.096 ms | 0 - 24 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.333 ms | 0 - 27 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 0.612 ms | 0 - 26 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.135 ms | 0 - 26 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.216 ms | 0 - 0 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.38 ms | 0 - 0 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.182 ms | 0 - 32 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 0.823 ms | 2 - 4 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 0.667 ms | 0 - 23 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.27 ms | 0 - 7 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.445 ms | 0 - 25 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 0.35 ms | 0 - 2 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 0.852 ms | 0 - 23 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.34 ms | 0 - 34 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Qualcomm® SA7255P | 0.667 ms | 0 - 23 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Qualcomm® SA8295P | 0.632 ms | 0 - 21 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.147 ms | 0 - 22 MB | NPU |
| MediaPipe-Selfie-Segmentation | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.263 ms | 0 - 23 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.317 ms | 0 - 29 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.459 ms | 0 - 35 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 1.818 ms | 0 - 25 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 0.69 ms | 0 - 2 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | float | Qualcomm® SA8775P | 1.028 ms | 0 - 27 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | float | Qualcomm® QCS9075 | 0.849 ms | 0 - 3 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 0.944 ms | 0 - 37 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | float | Qualcomm® SA7255P | 1.818 ms | 0 - 25 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | float | Qualcomm® SA8295P | 1.317 ms | 0 - 25 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.364 ms | 0 - 25 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.138 ms | 0 - 25 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.184 ms | 0 - 30 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Qualcomm® QCS6490 | 0.688 ms | 0 - 3 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 0.708 ms | 0 - 23 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.27 ms | 0 - 1 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Qualcomm® SA8775P | 0.467 ms | 0 - 24 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Qualcomm® QCS9075 | 0.365 ms | 0 - 2 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Qualcomm® QCM6690 | 0.836 ms | 0 - 21 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.352 ms | 0 - 32 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Qualcomm® SA7255P | 0.708 ms | 0 - 23 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Qualcomm® SA8295P | 0.614 ms | 0 - 21 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.145 ms | 0 - 23 MB | NPU |
| MediaPipe-Selfie-Segmentation | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.266 ms | 0 - 21 MB | NPU |
License
- The license for the original implementation of MediaPipe-Selfie-Segmentation 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.
