CVT: Optimized for Qualcomm Devices

Cross-View Transformer generates real-time bird's-eye view maps from multiple vehicle cameras for autonomous driving.

This is based on the implementation of CVT 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
TFLITE	float	Universal	QAIRT 2.45	Download

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

Model Details

Model Type: Model_use_case.driver_assistance

Model Stats:

• Inference latency: RealTime
• Input resolution: 1x6x3x224x480
• Model checkpoint: vehicles_50k.pt
• Model size (float): 5.18 MB
• Number of parameters: 1.33M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
CVT	ONNX	float	Snapdragon® X2 Elite	179.93 ms	8 - 8 MB	NPU
CVT	ONNX	float	Snapdragon® 8 Gen 3 Mobile	225.492 ms	8 - 2726 MB	NPU
CVT	ONNX	float	Snapdragon® 8 Gen 1 Mobile	440.517 ms	8 - 2756 MB	NPU
CVT	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	333.474 ms	0 - 23 MB	NPU
CVT	ONNX	float	Qualcomm® QCS8450	440.517 ms	8 - 2756 MB	NPU
CVT	ONNX	float	Qualcomm® Dragonwing™ Q-8750	188.235 ms	1 - 2394 MB	NPU
CVT	ONNX	float	Snapdragon® 8 Elite Mobile	188.235 ms	1 - 2394 MB	NPU
CVT	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	166.873 ms	0 - 2442 MB	NPU
CVT	QNN_DLC	float	Snapdragon® X2 Elite	182.316 ms	7 - 7 MB	NPU
CVT	QNN_DLC	float	Snapdragon® X Elite	286.424 ms	7 - 7 MB	NPU
CVT	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	224.915 ms	8 - 2574 MB	NPU
CVT	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	437.202 ms	7 - 2720 MB	NPU
CVT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	425.474 ms	2 - 2213 MB	NPU
CVT	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	333.437 ms	8 - 10 MB	NPU
CVT	QNN_DLC	float	Qualcomm® SA8775P	292.508 ms	2 - 2213 MB	NPU
CVT	QNN_DLC	float	Qualcomm® SA8650P	292.508 ms	2 - 2213 MB	NPU
CVT	QNN_DLC	float	Qualcomm® SA8255P	292.508 ms	2 - 2213 MB	NPU
CVT	QNN_DLC	float	Qualcomm® QCS8450	437.202 ms	7 - 2720 MB	NPU
CVT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	283.238 ms	7 - 17 MB	NPU
CVT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	286.424 ms	7 - 7 MB	NPU
CVT	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	184.434 ms	7 - 2297 MB	NPU
CVT	QNN_DLC	float	Qualcomm® SA7255P	425.474 ms	2 - 2213 MB	NPU
CVT	QNN_DLC	float	Qualcomm® SA8295P	333.241 ms	0 - 2324 MB	NPU
CVT	QNN_DLC	float	Snapdragon® 8 Elite Mobile	184.434 ms	7 - 2297 MB	NPU
CVT	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	167.54 ms	7 - 2318 MB	NPU
CVT	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	217.0 ms	89 - 2762 MB	NPU
CVT	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	420.379 ms	1 - 2728 MB	NPU
CVT	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	324.21 ms	0 - 4 MB	NPU
CVT	TFLITE	float	Qualcomm® QCS8450	420.379 ms	1 - 2728 MB	NPU
CVT	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	182.24 ms	0 - 2363 MB	NPU
CVT	TFLITE	float	Qualcomm® SA8295P	331.54 ms	0 - 2306 MB	NPU
CVT	TFLITE	float	Snapdragon® 8 Elite Mobile	182.24 ms	0 - 2363 MB	NPU
CVT	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	165.602 ms	0 - 2315 MB	NPU

License

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

References

• Cross-view Transformers for real-time Map-view Semantic Segmentation
• 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.