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 Mobile	QAIRT 2.42, ONNX Runtime 1.24.1	Download
PRECOMPILED_QNN_ONNX	float	Snapdragon® X2 Elite	QAIRT 2.42, ONNX Runtime 1.24.1	Download
PRECOMPILED_QNN_ONNX	float	Snapdragon® X Elite	QAIRT 2.42, ONNX Runtime 1.24.1	Download
PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Gen 3 Mobile	QAIRT 2.42, ONNX Runtime 1.24.1	Download
PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	QAIRT 2.42, ONNX Runtime 1.24.1	Download
QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Gen 5 Mobile	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	float	Snapdragon® X2 Elite	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	float	Snapdragon® X Elite	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	float	Snapdragon® 8 Gen 3 Mobile	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	float	Qualcomm® SA8775P	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite For Galaxy Mobile	QAIRT 2.43	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:

• Model checkpoint: camera-only-det.pth
• Input resolution: 1 x 6 x 3 x 256 x 704
• Number of parameters: 44M
• Model size: 171 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
BEVFusionDecoder	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	5.743 ms	16 - 26 MB	NPU
BEVFusionDecoder	PRECOMPILED_QNN_ONNX	float	Snapdragon® X2 Elite	7.013 ms	24 - 24 MB	NPU
BEVFusionDecoder	PRECOMPILED_QNN_ONNX	float	Snapdragon® X Elite	13.117 ms	23 - 23 MB	NPU
BEVFusionDecoder	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Gen 3 Mobile	9.851 ms	16 - 22 MB	NPU
BEVFusionDecoder	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	7.71 ms	12 - 19 MB	NPU
BEVFusionDecoder	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Gen 5 Mobile	5.714 ms	5 - 14 MB	NPU
BEVFusionDecoder	QNN_CONTEXT_BINARY	float	Snapdragon® X2 Elite	7.395 ms	5 - 5 MB	NPU
BEVFusionDecoder	QNN_CONTEXT_BINARY	float	Snapdragon® X Elite	13.106 ms	5 - 5 MB	NPU
BEVFusionDecoder	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Gen 3 Mobile	9.876 ms	5 - 12 MB	NPU
BEVFusionDecoder	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite For Galaxy Mobile	7.757 ms	0 - 9 MB	NPU
BEVFusionEncoder1	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	350.947 ms	51 - 61 MB	NPU
BEVFusionEncoder1	PRECOMPILED_QNN_ONNX	float	Snapdragon® X2 Elite	381.36 ms	102 - 102 MB	NPU
BEVFusionEncoder1	PRECOMPILED_QNN_ONNX	float	Snapdragon® X Elite	687.105 ms	101 - 101 MB	NPU
BEVFusionEncoder1	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Gen 3 Mobile	523.216 ms	32 - 44 MB	NPU
BEVFusionEncoder1	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	419.255 ms	37 - 49 MB	NPU
BEVFusionEncoder1	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Gen 5 Mobile	349.118 ms	13 - 22 MB	NPU
BEVFusionEncoder1	QNN_CONTEXT_BINARY	float	Snapdragon® X2 Elite	381.759 ms	12 - 12 MB	NPU
BEVFusionEncoder1	QNN_CONTEXT_BINARY	float	Snapdragon® X Elite	686.801 ms	12 - 12 MB	NPU
BEVFusionEncoder1	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Gen 3 Mobile	521.41 ms	13 - 20 MB	NPU
BEVFusionEncoder1	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite For Galaxy Mobile	420.68 ms	12 - 21 MB	NPU
BEVFusionEncoder2	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	2171.536 ms	293 - 303 MB	NPU
BEVFusionEncoder2	PRECOMPILED_QNN_ONNX	float	Snapdragon® X2 Elite	2354.007 ms	1058 - 1058 MB	NPU
BEVFusionEncoder2	PRECOMPILED_QNN_ONNX	float	Snapdragon® X Elite	3430.112 ms	1058 - 1058 MB	NPU
BEVFusionEncoder2	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Gen 3 Mobile	2747.846 ms	538 - 545 MB	NPU
BEVFusionEncoder2	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	2464.98 ms	512 - 519 MB	NPU
BEVFusionEncoder2	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Gen 5 Mobile	2172.187 ms	16 - 26 MB	NPU
BEVFusionEncoder2	QNN_CONTEXT_BINARY	float	Snapdragon® X2 Elite	2429.811 ms	17 - 17 MB	NPU
BEVFusionEncoder2	QNN_CONTEXT_BINARY	float	Snapdragon® X Elite	3530.924 ms	17 - 17 MB	NPU
BEVFusionEncoder2	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Gen 3 Mobile	2734.547 ms	17 - 27 MB	NPU
BEVFusionEncoder2	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite For Galaxy Mobile	2451.822 ms	17 - 30 MB	NPU
BEVFusionEncoder3	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	369.72 ms	508 - 518 MB	NPU
BEVFusionEncoder3	PRECOMPILED_QNN_ONNX	float	Snapdragon® X2 Elite	355.989 ms	610 - 610 MB	NPU
BEVFusionEncoder3	PRECOMPILED_QNN_ONNX	float	Snapdragon® X Elite	700.88 ms	610 - 610 MB	NPU
BEVFusionEncoder3	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Gen 3 Mobile	582.578 ms	572 - 582 MB	NPU
BEVFusionEncoder3	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	499.155 ms	571 - 583 MB	NPU
BEVFusionEncoder3	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Gen 5 Mobile	368.343 ms	609 - 619 MB	NPU
BEVFusionEncoder3	QNN_CONTEXT_BINARY	float	Snapdragon® X2 Elite	356.194 ms	610 - 610 MB	NPU
BEVFusionEncoder3	QNN_CONTEXT_BINARY	float	Snapdragon® X Elite	700.168 ms	610 - 610 MB	NPU
BEVFusionEncoder3	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Gen 3 Mobile	591.853 ms	623 - 635 MB	NPU
BEVFusionEncoder3	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite For Galaxy Mobile	500.811 ms	608 - 623 MB	NPU
BEVFusionEncoder4	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	6.771 ms	30 - 40 MB	NPU
BEVFusionEncoder4	PRECOMPILED_QNN_ONNX	float	Snapdragon® X2 Elite	7.929 ms	14 - 14 MB	NPU
BEVFusionEncoder4	PRECOMPILED_QNN_ONNX	float	Snapdragon® X Elite	12.165 ms	18 - 18 MB	NPU
BEVFusionEncoder4	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Gen 3 Mobile	9.159 ms	31 - 38 MB	NPU
BEVFusionEncoder4	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	7.707 ms	14 - 21 MB	NPU
BEVFusionEncoder4	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Gen 5 Mobile	6.783 ms	18 - 28 MB	NPU
BEVFusionEncoder4	QNN_CONTEXT_BINARY	float	Snapdragon® X2 Elite	8.702 ms	19 - 19 MB	NPU
BEVFusionEncoder4	QNN_CONTEXT_BINARY	float	Snapdragon® X Elite	11.936 ms	19 - 19 MB	NPU
BEVFusionEncoder4	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Gen 3 Mobile	9.085 ms	19 - 26 MB	NPU
BEVFusionEncoder4	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite For Galaxy Mobile	7.826 ms	18 - 27 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.