CenterNet-3D: Optimized for Qualcomm Devices

CenterNet 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 CenterNet-3D 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
QNN_CONTEXT_BINARY	w8a16	qualcomm_qcs8450_proxy	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	w8a16	qualcomm_qcs8550_proxy	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	w8a16	qualcomm_qcs9075	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	w8a16	qualcomm_sa7255p	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	w8a16	qualcomm_sa8295p	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	w8a16	qualcomm_sa8775p	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	w8a16	qualcomm_snapdragon_8_elite_for_galaxy	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	w8a16	qualcomm_snapdragon_8gen3	QAIRT 2.43	Download
QNN_CONTEXT_BINARY	w8a16	qualcomm_snapdragon_x_elite	QAIRT 2.43	Download

For more device-specific assets and performance metrics, visit CenterNet-3D 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 CenterNet-3D on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.driver_assistance

Model Stats:

• Model checkpoint: ddd_3dop.pth
• Input resolution: 1 x 3 x 384 x 1280
• Number of parameters: 20.6M
• Model size: 79 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
CenterNet-3D	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	2749.356 ms	3 - 13 MB	NPU
CenterNet-3D	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® X2 Elite	1959.103 ms	73 - 73 MB	NPU
CenterNet-3D	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® X Elite	2698.898 ms	55 - 55 MB	NPU
CenterNet-3D	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	2317.181 ms	5 - 11 MB	NPU
CenterNet-3D	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	3001.658 ms	1 - 63 MB	NPU
CenterNet-3D	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCS9075	2872.552 ms	0 - 6 MB	NPU
CenterNet-3D	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	2374.057 ms	1 - 8 MB	NPU
CenterNet-3D	QNN_CONTEXT_BINARY	w8a16	Snapdragon® X Elite	2745.583 ms	3 - 3 MB	NPU
CenterNet-3D	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Gen 3 Mobile	2360.5 ms	3 - 12 MB	NPU
CenterNet-3D	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS8275 (Proxy)	3937.785 ms	1 - 9 MB	NPU
CenterNet-3D	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS8550 (Proxy)	3073.546 ms	3 - 5 MB	NPU
CenterNet-3D	QNN_CONTEXT_BINARY	w8a16	Qualcomm® SA8775P	2972.172 ms	1 - 9 MB	NPU
CenterNet-3D	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS9075	2934.794 ms	5 - 10 MB	NPU
CenterNet-3D	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS8450 (Proxy)	4429.015 ms	7 - 16 MB	NPU
CenterNet-3D	QNN_CONTEXT_BINARY	w8a16	Qualcomm® SA7255P	3937.785 ms	1 - 9 MB	NPU
CenterNet-3D	QNN_CONTEXT_BINARY	w8a16	Qualcomm® SA8295P	3507.662 ms	2 - 7 MB	NPU
CenterNet-3D	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	2407.793 ms	3 - 16 MB	NPU

License

• The license for the original implementation of CenterNet-3D can be found [here](https://github.com/xingyizhou/CenterNet/blob/master/LICENSE https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf).

References

• Objects as Points
• 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.