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	ONNX Runtime 1.25.0	Download
TFLITE	float	Universal		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:

• Model checkpoint: bevdet-r50.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
BEVDet	ONNX	float	Snapdragon® X2 Elite	591.673 ms	736 - 736 MB	CPU
BEVDet	ONNX	float	Snapdragon® X Elite	2720.251 ms	466 - 466 MB	CPU
BEVDet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	2255.774 ms	213 - 222 MB	CPU
BEVDet	ONNX	float	Snapdragon® 8 Gen 1 Mobile	2961.223 ms	204 - 220 MB	CPU
BEVDet	ONNX	float	Qualcomm® QCS8550 (Proxy)	2850.303 ms	187 - 189 MB	CPU
BEVDet	ONNX	float	Qualcomm® QCS8450	2961.223 ms	204 - 220 MB	CPU
BEVDet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1229.663 ms	238 - 246 MB	CPU
BEVDet	ONNX	float	Qualcomm® QCS9075	1506.065 ms	241 - 255 MB	CPU
BEVDet	ONNX	float	Snapdragon® 8 Elite Mobile	1419.093 ms	237 - 246 MB	CPU
BEVDet	ONNX	float	Qualcomm® QCS8750	1419.093 ms	237 - 246 MB	CPU
BEVDet	ONNX	float	Qualcomm® QCS7181	2720.251 ms	466 - 466 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Snapdragon® X2 Elite	819.326 ms	712 - 712 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Snapdragon® X Elite	4253.027 ms	709 - 709 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Snapdragon® 8 Gen 3 Mobile	2438.833 ms	360 - 373 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Qualcomm® QCS8550 (Proxy)	2688.828 ms	389 - 402 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Qualcomm® QCS9075	1869.81 ms	423 - 433 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Qualcomm® QCS7181	4253.027 ms	709 - 709 MB	CPU
BEVDet	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1943.327 ms	121 - 137 MB	CPU
BEVDet	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	2932.543 ms	128 - 146 MB	CPU
BEVDet	TFLITE	float	Qualcomm® QCS8275	3159.154 ms	128 - 137 MB	CPU
BEVDet	TFLITE	float	Qualcomm® QCS8550 (Proxy)	2325.841 ms	127 - 130 MB	CPU
BEVDet	TFLITE	float	Qualcomm® SA8775P	2485.313 ms	128 - 138 MB	CPU
BEVDet	TFLITE	float	Qualcomm® SA8650P	2485.313 ms	128 - 138 MB	CPU
BEVDet	TFLITE	float	Qualcomm® SA8255P	2485.313 ms	128 - 138 MB	CPU
BEVDet	TFLITE	float	Qualcomm® QCS8450	2932.543 ms	128 - 146 MB	CPU
BEVDet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	385.284 ms	78 - 88 MB	CPU
BEVDet	TFLITE	float	Qualcomm® SA7255P	3159.154 ms	128 - 137 MB	CPU
BEVDet	TFLITE	float	Qualcomm® QCS9075	2392.932 ms	126 - 1330 MB	CPU
BEVDet	TFLITE	float	Snapdragon® 8 Elite Mobile	1251.374 ms	78 - 88 MB	CPU
BEVDet	TFLITE	float	Qualcomm® SA8295P	1835.895 ms	128 - 138 MB	CPU
BEVDet	TFLITE	float	Qualcomm® QCS8750	1251.374 ms	78 - 88 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

• 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.