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 1.24.1	Download
ONNX	w8a16_mixed_fp16	Universal	ONNX Runtime 1.24.1	Download
TFLITE	float	Universal	TFLite 2.17.0	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® X Elite	720.637 ms	733 - 733 MB	CPU
BEVDet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	2185.974 ms	210 - 220 MB	CPU
BEVDet	ONNX	float	Qualcomm® QCS8550 (Proxy)	2548.327 ms	183 - 188 MB	CPU
BEVDet	ONNX	float	Qualcomm® QCS9075	1515.86 ms	237 - 251 MB	CPU
BEVDet	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	1322.478 ms	239 - 251 MB	CPU
BEVDet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1508.474 ms	248 - 258 MB	CPU
BEVDet	ONNX	float	Snapdragon® X2 Elite	584.082 ms	736 - 736 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Snapdragon® X Elite	855.454 ms	1238 - 1238 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Snapdragon® 8 Gen 3 Mobile	2507.491 ms	359 - 373 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Qualcomm® QCS8550 (Proxy)	2760.446 ms	390 - 403 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Qualcomm® QCS9075	1856.516 ms	423 - 433 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Snapdragon® 8 Elite For Galaxy Mobile	1631.784 ms	327 - 341 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Snapdragon® 8 Elite Gen 5 Mobile	1486.873 ms	320 - 334 MB	CPU
BEVDet	ONNX	w8a16_mixed_fp16	Snapdragon® X2 Elite	820.509 ms	713 - 713 MB	CPU
BEVDet	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1982.681 ms	127 - 144 MB	CPU
BEVDet	TFLITE	float	Qualcomm® QCS8275 (Proxy)	3171.244 ms	129 - 138 MB	CPU
BEVDet	TFLITE	float	Qualcomm® QCS8550 (Proxy)	2053.845 ms	104 - 107 MB	CPU
BEVDet	TFLITE	float	Qualcomm® SA8775P	1416.041 ms	120 - 415 MB	CPU
BEVDet	TFLITE	float	Qualcomm® QCS9075	2427.345 ms	126 - 1473 MB	CPU
BEVDet	TFLITE	float	Qualcomm® QCS8450 (Proxy)	2614.274 ms	127 - 145 MB	CPU
BEVDet	TFLITE	float	Qualcomm® SA7255P	3171.244 ms	129 - 138 MB	CPU
BEVDet	TFLITE	float	Qualcomm® SA8295P	1939.814 ms	203 - 211 MB	CPU
BEVDet	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	1260.614 ms	107 - 120 MB	CPU
BEVDet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1066.862 ms	90 - 101 MB	CPU

License

• The license for the original implementation of BEVDet can be found [here](https://github.com/HuangJunJie2017/BEVDet/blob/dev3.0/LICENSE https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf).

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.