BEVFormer: Optimized for Qualcomm Devices

Bevformer is a SOTA model of interest to the Auto BU.

This is based on the implementation of BEVFormer 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

For more device-specific assets and performance metrics, visit BEVFormer 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 BEVFormer on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.driver_assistance

Model Stats:

• Input preprocessing: Images must be ImageNet mean/std normalized before inference; the model does not normalize internally.
• Input resolution: 6 x 3 x 480 x 800
• Model checkpoint: bevformer_tiny_deformable_optimized_exp_86_epoch_24.pth
• Model size: 120 MB
• Number of parameters: 27M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
BEVFormer	ONNX	float	Snapdragon® X2 Elite	1277.286 ms	29 - 29 MB	NPU
BEVFormer	ONNX	float	Snapdragon® X Elite	2394.933 ms	51 - 51 MB	NPU
BEVFormer	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1393.623 ms	0 - 1124 MB	NPU
BEVFormer	ONNX	float	Snapdragon® 8 Gen 1 Mobile	2236.206 ms	26 - 1207 MB	NPU
BEVFormer	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	2320.09 ms	29 - 61 MB	NPU
BEVFormer	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1819.952 ms	0 - 53 MB	NPU
BEVFormer	ONNX	float	Qualcomm® QCS8450	2236.206 ms	26 - 1207 MB	NPU
BEVFormer	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	1961.668 ms	29 - 60 MB	NPU
BEVFormer	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	2394.933 ms	51 - 51 MB	NPU
BEVFormer	ONNX	float	Qualcomm® Dragonwing™ Q-8750	1352.307 ms	2 - 843 MB	NPU
BEVFormer	ONNX	float	Snapdragon® 8 Elite Mobile	1352.307 ms	2 - 843 MB	NPU
BEVFormer	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1223.467 ms	0 - 860 MB	NPU
BEVFormer	QNN_DLC	float	Snapdragon® X2 Elite	1295.087 ms	29 - 29 MB	NPU
BEVFormer	QNN_DLC	float	Snapdragon® X Elite	1946.029 ms	29 - 29 MB	NPU
BEVFormer	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1667.489 ms	29 - 1126 MB	NPU
BEVFormer	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	1992.252 ms	4 - 1097 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	2347.891 ms	29 - 62 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1880.741 ms	29 - 34 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® SA8775P	2011.958 ms	26 - 863 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® SA8650P	2011.958 ms	26 - 863 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® SA8255P	2011.958 ms	26 - 863 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® QCS8450	1992.252 ms	4 - 1097 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	2013.791 ms	29 - 62 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	1946.029 ms	29 - 29 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	1266.836 ms	26 - 913 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® SA7255P	3964.374 ms	26 - 855 MB	NPU
BEVFormer	QNN_DLC	float	Qualcomm® SA8295P	2190.604 ms	26 - 865 MB	NPU
BEVFormer	QNN_DLC	float	Snapdragon® 8 Elite Mobile	1266.836 ms	26 - 913 MB	NPU
BEVFormer	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	1231.303 ms	14 - 870 MB	NPU

License

• The license for the original implementation of BEVFormer can be found here.

References

• BEVFormer: Learning Bird's-Eye-View Representation from Multi-Camera Images via Spatiotemporal Transformers
• 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.