BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's-Eye View Representation
Paper • 2205.13542 • Published
BEVFusion: Optimized for Qualcomm Devices
BeVFusion 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 BEVFusion 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 |
|---|---|---|---|---|
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.50 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.50 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | QAIRT 2.50 | Download |
For more device-specific assets and performance metrics, visit BEVFusion 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 BEVFusion 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: camera-only-det.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 | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 5.791 ms | 12 - 19 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.697 ms | 11 - 21 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 6.925 ms | 22 - 22 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 13.071 ms | 24 - 24 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 9.893 ms | 15 - 23 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 13.071 ms | 24 - 24 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 7.697 ms | 11 - 21 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 5.796 ms | 0 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.627 ms | 0 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 7.455 ms | 5 - 5 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 13.259 ms | 5 - 5 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 9.868 ms | 5 - 13 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8650P | 23.802 ms | 4 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8255P | 23.802 ms | 4 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-X7181 | 13.259 ms | 5 - 5 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ Q-8750 | 7.627 ms | 0 - 9 MB | NPU |
| encoder_1 | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 367.542 ms | 38 - 51 MB | NPU |
| encoder_1 | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 449.719 ms | 38 - 49 MB | NPU |
| encoder_1 | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 401.23 ms | 66 - 66 MB | NPU |
| encoder_1 | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 731.32 ms | 86 - 86 MB | NPU |
| encoder_1 | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 552.897 ms | 39 - 51 MB | NPU |
| encoder_1 | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 731.32 ms | 86 - 86 MB | NPU |
| encoder_1 | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 449.719 ms | 38 - 49 MB | NPU |
| encoder_1 | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 368.77 ms | 2 - 11 MB | NPU |
| encoder_1 | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 449.214 ms | 12 - 20 MB | NPU |
| encoder_1 | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 400.306 ms | 12 - 12 MB | NPU |
| encoder_1 | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 708.492 ms | 12 - 12 MB | NPU |
| encoder_1 | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 553.109 ms | 13 - 19 MB | NPU |
| encoder_1 | QNN_CONTEXT_BINARY | float | Qualcomm® SA8650P | 720.121 ms | 0 - 9 MB | NPU |
| encoder_1 | QNN_CONTEXT_BINARY | float | Qualcomm® SA8255P | 720.121 ms | 0 - 9 MB | NPU |
| encoder_1 | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-X7181 | 708.492 ms | 12 - 12 MB | NPU |
| encoder_1 | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ Q-8750 | 449.214 ms | 12 - 20 MB | NPU |
| encoder_2 | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 2266.532 ms | 492 - 499 MB | NPU |
| encoder_2 | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 2674.917 ms | 290 - 300 MB | NPU |
| encoder_2 | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 2434.991 ms | 1058 - 1058 MB | NPU |
| encoder_2 | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 3421.274 ms | 1058 - 1058 MB | NPU |
| encoder_2 | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 2887.244 ms | 600 - 607 MB | NPU |
| encoder_2 | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 3421.274 ms | 1058 - 1058 MB | NPU |
| encoder_2 | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 2674.917 ms | 290 - 300 MB | NPU |
| encoder_2 | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 2255.302 ms | 17 - 28 MB | NPU |
| encoder_2 | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 2691.979 ms | 1 - 14 MB | NPU |
| encoder_2 | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 2436.759 ms | 17 - 17 MB | NPU |
| encoder_2 | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 3507.936 ms | 17 - 17 MB | NPU |
| encoder_2 | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 2930.026 ms | 17 - 29 MB | NPU |
| encoder_2 | QNN_CONTEXT_BINARY | float | Qualcomm® SA8650P | 3459.164 ms | 2 - 11 MB | NPU |
| encoder_2 | QNN_CONTEXT_BINARY | float | Qualcomm® SA8255P | 3459.164 ms | 2 - 11 MB | NPU |
| encoder_2 | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-X7181 | 3507.936 ms | 17 - 17 MB | NPU |
| encoder_2 | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ Q-8750 | 2691.979 ms | 1 - 14 MB | NPU |
| encoder_3 | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 6.675 ms | 15 - 24 MB | NPU |
| encoder_3 | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.827 ms | 12 - 23 MB | NPU |
| encoder_3 | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 7.922 ms | 19 - 19 MB | NPU |
| encoder_3 | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 11.799 ms | 18 - 18 MB | NPU |
| encoder_3 | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 9.343 ms | 31 - 39 MB | NPU |
| encoder_3 | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 11.799 ms | 18 - 18 MB | NPU |
| encoder_3 | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 7.827 ms | 12 - 23 MB | NPU |
| encoder_3 | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 6.803 ms | 16 - 24 MB | NPU |
| encoder_3 | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.783 ms | 18 - 26 MB | NPU |
| encoder_3 | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 8.522 ms | 19 - 19 MB | NPU |
| encoder_3 | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 12.202 ms | 19 - 19 MB | NPU |
| encoder_3 | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 9.251 ms | 18 - 26 MB | NPU |
| encoder_3 | QNN_CONTEXT_BINARY | float | Qualcomm® SA8650P | 15.076 ms | 18 - 23 MB | NPU |
| encoder_3 | QNN_CONTEXT_BINARY | float | Qualcomm® SA8255P | 15.076 ms | 18 - 23 MB | NPU |
| encoder_3 | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-X7181 | 12.202 ms | 19 - 19 MB | NPU |
| encoder_3 | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ Q-8750 | 7.783 ms | 18 - 26 MB | NPU |
License
- The license for the original implementation of BEVFusion can be found here.
References
- BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's-Eye View Representation
- Source Model Implementation
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.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support