
BEVDet: Optimized for Qualcomm Devices
BEVDet is a machine learning model for generating a birds eye view represenation from the sensors(cameras) mounted on a vehicle.
This is based on the implementation of BEVDet 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 2.45, ONNX Runtime 1.27.1 | Download |
| QNN_DLC | float | Universal | QAIRT 2.45 | Download |
| TFLITE | float | Universal | QAIRT 2.45 | Download |
For more device-specific assets and performance metrics, visit BEVDet 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 BEVDet on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.driver_assistance
Model Stats:
- Input resolution: 1 x 6 x 3 x 256 x 704
- Model checkpoint: bevdet-r50.pth
- Model size: 171 MB
- Number of parameters: 44M
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| decoder | ONNX | float | Snapdragon® X2 Elite | 6.436 ms | 4 - 4 MB | NPU |
| decoder | ONNX | float | Snapdragon® X Elite | 12.421 ms | 34 - 34 MB | NPU |
| decoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 8.827 ms | 0 - 174 MB | NPU |
| decoder | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 23.547 ms | 0 - 170 MB | NPU |
| decoder | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 25.915 ms | 4 - 11 MB | NPU |
| decoder | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 11.95 ms | 0 - 37 MB | NPU |
| decoder | ONNX | float | Qualcomm® QCS8450 | 23.547 ms | 0 - 170 MB | NPU |
| decoder | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 23.204 ms | 4 - 11 MB | NPU |
| decoder | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 12.421 ms | 34 - 34 MB | NPU |
| decoder | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 7.587 ms | 3 - 150 MB | NPU |
| decoder | ONNX | float | Snapdragon® 8 Elite Mobile | 7.587 ms | 3 - 150 MB | NPU |
| decoder | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.472 ms | 3 - 154 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® X2 Elite | 6.757 ms | 4 - 4 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® X Elite | 12.556 ms | 4 - 4 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 8.787 ms | 0 - 170 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 23.203 ms | 0 - 169 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 25.555 ms | 4 - 12 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 11.863 ms | 4 - 6 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® SA8775P | 23.121 ms | 1 - 148 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® SA8650P | 23.121 ms | 1 - 148 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® SA8255P | 23.121 ms | 1 - 148 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® QCS8450 | 23.203 ms | 0 - 169 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 23.468 ms | 4 - 11 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 12.556 ms | 4 - 4 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 7.56 ms | 0 - 148 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® SA7255P | 87.64 ms | 1 - 145 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® SA8295P | 25.535 ms | 0 - 142 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 7.56 ms | 0 - 148 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.567 ms | 4 - 152 MB | NPU |
| decoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 8.828 ms | 1 - 207 MB | NPU |
| decoder | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 23.206 ms | 1 - 206 MB | NPU |
| decoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 25.656 ms | 1 - 46 MB | NPU |
| decoder | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 11.971 ms | 1 - 3 MB | NPU |
| decoder | TFLITE | float | Qualcomm® SA8775P | 23.126 ms | 1 - 153 MB | NPU |
| decoder | TFLITE | float | Qualcomm® SA8650P | 23.126 ms | 1 - 153 MB | NPU |
| decoder | TFLITE | float | Qualcomm® SA8255P | 23.126 ms | 1 - 153 MB | NPU |
| decoder | TFLITE | float | Qualcomm® QCS8450 | 23.206 ms | 1 - 206 MB | NPU |
| decoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 23.89 ms | 1 - 46 MB | NPU |
| decoder | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 7.54 ms | 0 - 148 MB | NPU |
| decoder | TFLITE | float | Qualcomm® SA7255P | 87.582 ms | 1 - 149 MB | NPU |
| decoder | TFLITE | float | Qualcomm® SA8295P | 25.532 ms | 1 - 146 MB | NPU |
| decoder | TFLITE | float | Snapdragon® 8 Elite Mobile | 7.54 ms | 0 - 148 MB | NPU |
| decoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.472 ms | 1 - 155 MB | NPU |
| encoder | ONNX | float | Snapdragon® X2 Elite | 23.458 ms | 12 - 12 MB | NPU |
| encoder | ONNX | float | Snapdragon® X Elite | 44.747 ms | 39 - 39 MB | NPU |
| encoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 36.861 ms | 0 - 420 MB | NPU |
| encoder | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 82.171 ms | 1 - 328 MB | NPU |
| encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 72.746 ms | 12 - 28 MB | NPU |
| encoder | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 46.745 ms | 0 - 45 MB | NPU |
| encoder | ONNX | float | Qualcomm® QCS8450 | 82.171 ms | 1 - 328 MB | NPU |
| encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 67.322 ms | 12 - 27 MB | NPU |
| encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 44.747 ms | 39 - 39 MB | NPU |
| encoder | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 26.679 ms | 6 - 321 MB | NPU |
| encoder | ONNX | float | Snapdragon® 8 Elite Mobile | 26.679 ms | 6 - 321 MB | NPU |
| encoder | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 21.145 ms | 6 - 337 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® X2 Elite | 22.822 ms | 12 - 12 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® X Elite | 44.003 ms | 12 - 12 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 37.315 ms | 12 - 427 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 81.348 ms | 4 - 324 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 67.927 ms | 12 - 29 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 44.045 ms | 13 - 414 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® SA8775P | 64.989 ms | 0 - 306 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® SA8650P | 64.989 ms | 0 - 306 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® SA8255P | 64.989 ms | 0 - 306 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® QCS8450 | 81.348 ms | 4 - 324 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 65.05 ms | 12 - 28 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 44.003 ms | 12 - 12 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 25.259 ms | 12 - 328 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® SA7255P | 207.81 ms | 1 - 306 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® SA8295P | 68.789 ms | 0 - 231 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 25.259 ms | 12 - 328 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 20.119 ms | 12 - 349 MB | NPU |
| encoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 35.917 ms | 24 - 477 MB | NPU |
| encoder | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 76.238 ms | 25 - 391 MB | NPU |
| encoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 73.236 ms | 25 - 115 MB | NPU |
| encoder | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 45.258 ms | 25 - 27 MB | NPU |
| encoder | TFLITE | float | Qualcomm® SA8775P | 67.313 ms | 25 - 353 MB | NPU |
| encoder | TFLITE | float | Qualcomm® SA8650P | 67.313 ms | 25 - 353 MB | NPU |
| encoder | TFLITE | float | Qualcomm® SA8255P | 67.313 ms | 25 - 353 MB | NPU |
| encoder | TFLITE | float | Qualcomm® QCS8450 | 76.238 ms | 25 - 391 MB | NPU |
| encoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 68.202 ms | 23 - 112 MB | NPU |
| encoder | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 28.858 ms | 24 - 359 MB | NPU |
| encoder | TFLITE | float | Qualcomm® SA7255P | 213.598 ms | 25 - 352 MB | NPU |
| encoder | TFLITE | float | Qualcomm® SA8295P | 67.796 ms | 25 - 269 MB | NPU |
| encoder | TFLITE | float | Snapdragon® 8 Elite Mobile | 28.858 ms | 24 - 359 MB | NPU |
| encoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 21.287 ms | 24 - 365 MB | NPU |
| pooler | ONNX | float | Snapdragon® X2 Elite | 17.265 ms | 284 - 284 MB | CPU |
| pooler | ONNX | float | Snapdragon® X Elite | 78.002 ms | 283 - 283 MB | CPU |
| pooler | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 24.753 ms | 278 - 287 MB | CPU |
| pooler | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 33.964 ms | 257 - 268 MB | CPU |
| pooler | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 54.586 ms | 274 - 278 MB | CPU |
| pooler | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 26.623 ms | 274 - 276 MB | CPU |
| pooler | ONNX | float | Qualcomm® QCS8450 | 33.964 ms | 257 - 268 MB | CPU |
| pooler | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 47.101 ms | 277 - 281 MB | CPU |
| pooler | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 78.002 ms | 283 - 283 MB | CPU |
| pooler | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 21.576 ms | 279 - 291 MB | CPU |
| pooler | ONNX | float | Snapdragon® 8 Elite Mobile | 21.576 ms | 279 - 291 MB | CPU |
| pooler | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 18.256 ms | 278 - 286 MB | CPU |
| pooler | QNN_DLC | float | Snapdragon® X2 Elite | 322.292 ms | 4 - 4 MB | CPU |
| pooler | QNN_DLC | float | Snapdragon® X Elite | 332.112 ms | 4 - 4 MB | CPU |
| pooler | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 149.93 ms | 4 - 14 MB | CPU |
| pooler | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 200.167 ms | 4 - 20 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 369.409 ms | 268 - 736 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 166.039 ms | 5 - 499 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® SA8775P | 308.633 ms | 4 - 17 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® SA8650P | 308.633 ms | 4 - 17 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® SA8255P | 308.633 ms | 4 - 17 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® QCS8450 | 200.167 ms | 4 - 20 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 327.313 ms | 269 - 737 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 332.112 ms | 4 - 4 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 173.973 ms | 4 - 17 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® SA7255P | 358.012 ms | 4 - 16 MB | CPU |
| pooler | QNN_DLC | float | Qualcomm® SA8295P | 306.432 ms | 4 - 16 MB | CPU |
| pooler | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 173.973 ms | 4 - 17 MB | CPU |
| pooler | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 215.454 ms | 4 - 13 MB | CPU |
| pooler | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 46.741 ms | 127 - 140 MB | CPU |
| pooler | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 74.48 ms | 120 - 134 MB | CPU |
| pooler | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 127.141 ms | 127 - 321 MB | CPU |
| pooler | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 64.365 ms | 128 - 130 MB | CPU |
| pooler | TFLITE | float | Qualcomm® SA8775P | 90.951 ms | 130 - 136 MB | CPU |
| pooler | TFLITE | float | Qualcomm® SA8650P | 90.951 ms | 130 - 136 MB | CPU |
| pooler | TFLITE | float | Qualcomm® SA8255P | 90.951 ms | 130 - 136 MB | CPU |
| pooler | TFLITE | float | Qualcomm® QCS8450 | 74.48 ms | 120 - 134 MB | CPU |
| pooler | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 106.262 ms | 127 - 320 MB | CPU |
| pooler | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 52.861 ms | 126 - 138 MB | CPU |
| pooler | TFLITE | float | Qualcomm® SA7255P | 108.327 ms | 128 - 137 MB | CPU |
| pooler | TFLITE | float | Qualcomm® SA8295P | 86.391 ms | 127 - 133 MB | CPU |
| pooler | TFLITE | float | Snapdragon® 8 Elite Mobile | 52.861 ms | 126 - 138 MB | CPU |
| pooler | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 46.184 ms | 127 - 139 MB | CPU |
License
- The license for the original implementation of BEVDet can be found here.
References
- BEVDet: High-Performance Multi-Camera 3D Object Detection in Bird-Eye-View
- Source Model Implementation
Community
- For questions or feedback, chat with our engineers in Qualcomm® AI Hub Workbench.