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 2.50, ONNX Runtime 1.30.0	Download
ONNX	w8a16_mixed_int16	Universal	QAIRT 2.50, ONNX Runtime 1.30.0	Download
QNN_DLC	float	Universal	QAIRT 2.50	Download
QNN_DLC	w8a16_mixed_int16	Universal	QAIRT 2.50	Download
TFLITE	float	Universal	QAIRT 2.50	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® 8 Elite Gen 5 For Galaxy Mobile	159.12 ms	0 - 2225 MB	NPU
CVT	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	180.899 ms	1 - 2177 MB	NPU
CVT	ONNX	float	Snapdragon® X2 Elite	169.867 ms	8 - 8 MB	NPU
CVT	ONNX	float	Snapdragon® X Elite	263.565 ms	18 - 18 MB	NPU
CVT	ONNX	float	Snapdragon® 8 Gen 3 Mobile	214.71 ms	8 - 2497 MB	NPU
CVT	ONNX	float	Snapdragon® 8 Gen 1 Mobile	358.64 ms	8 - 2652 MB	NPU
CVT	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	247.267 ms	7 - 11 MB	NPU
CVT	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	316.396 ms	0 - 20 MB	NPU
CVT	ONNX	float	Qualcomm® QCS8450	358.64 ms	8 - 2652 MB	NPU
CVT	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	279.276 ms	7 - 11 MB	NPU
CVT	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	263.565 ms	18 - 18 MB	NPU
CVT	ONNX	float	Qualcomm® Dragonwing™ Q-8750	180.899 ms	1 - 2177 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	183.166 ms	1 - 2322 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Snapdragon® 8 Elite For Galaxy Mobile	202.07 ms	3 - 2284 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Snapdragon® X2 Elite	197.071 ms	4 - 4 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Snapdragon® X Elite	299.876 ms	22 - 22 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Snapdragon® 8 Gen 3 Mobile	236.557 ms	4 - 2552 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Snapdragon® 8 Gen 1 Mobile	357.38 ms	4 - 2580 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Qualcomm® Dragonwing™ IQ-8275	271.591 ms	4 - 8 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	338.842 ms	0 - 1444 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Qualcomm® QCS8450	357.38 ms	4 - 2580 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Qualcomm® Dragonwing™ IQ-9075	296.18 ms	4 - 7 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Qualcomm® Dragonwing™ IQ-X7181	299.876 ms	22 - 22 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Qualcomm® Dragonwing™ Q-7790	1359.819 ms	67 - 837 MB	NPU
CVT	ONNX	w8a16_mixed_int16	Qualcomm® Dragonwing™ Q-8750	202.07 ms	3 - 2284 MB	NPU
CVT	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	163.935 ms	7 - 2195 MB	NPU
CVT	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	182.742 ms	7 - 2080 MB	NPU
CVT	QNN_DLC	float	Snapdragon® X2 Elite	177.399 ms	7 - 7 MB	NPU
CVT	QNN_DLC	float	Snapdragon® X Elite	262.543 ms	7 - 7 MB	NPU
CVT	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	217.997 ms	5 - 2461 MB	NPU
CVT	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	369.058 ms	5 - 2608 MB	NPU
CVT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	250.144 ms	7 - 18 MB	NPU
CVT	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	318.158 ms	8 - 12 MB	NPU
CVT	QNN_DLC	float	Qualcomm® QCS8450	369.058 ms	5 - 2608 MB	NPU
CVT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	282.284 ms	7 - 17 MB	NPU
CVT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	262.543 ms	7 - 7 MB	NPU
CVT	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	182.742 ms	7 - 2080 MB	NPU
CVT	QNN_DLC	float	Qualcomm® SA8295P	321.101 ms	1 - 2252 MB	NPU
CVT	QNN_DLC	w8a16_mixed_int16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	191.997 ms	4 - 2269 MB	NPU
CVT	QNN_DLC	w8a16_mixed_int16	Snapdragon® 8 Elite For Galaxy Mobile	211.238 ms	4 - 2261 MB	NPU
CVT	QNN_DLC	w8a16_mixed_int16	Snapdragon® X2 Elite	201.983 ms	4 - 4 MB	NPU
CVT	QNN_DLC	w8a16_mixed_int16	Snapdragon® X Elite	308.482 ms	4 - 4 MB	NPU
CVT	QNN_DLC	w8a16_mixed_int16	Snapdragon® 8 Gen 3 Mobile	243.781 ms	5 - 2486 MB	NPU
CVT	QNN_DLC	w8a16_mixed_int16	Qualcomm® Dragonwing™ IQ-8275	277.829 ms	4 - 10 MB	NPU
CVT	QNN_DLC	w8a16_mixed_int16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	341.113 ms	4 - 7 MB	NPU
CVT	QNN_DLC	w8a16_mixed_int16	Qualcomm® Dragonwing™ IQ-9075	304.259 ms	4 - 10 MB	NPU
CVT	QNN_DLC	w8a16_mixed_int16	Qualcomm® Dragonwing™ IQ-X7181	308.482 ms	4 - 4 MB	NPU
CVT	QNN_DLC	w8a16_mixed_int16	Qualcomm® Dragonwing™ Q-8750	211.238 ms	4 - 2261 MB	NPU
CVT	TFLITE	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	159.355 ms	0 - 2197 MB	NPU
CVT	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	179.294 ms	1 - 2095 MB	NPU
CVT	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	214.275 ms	1 - 2502 MB	NPU
CVT	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	330.218 ms	0 - 2643 MB	NPU
CVT	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	247.005 ms	0 - 34 MB	NPU
CVT	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	314.181 ms	0 - 1683 MB	NPU
CVT	TFLITE	float	Qualcomm® SA8650P	571.046 ms	48 - 61 MB	CPU
CVT	TFLITE	float	Qualcomm® SA8255P	571.046 ms	48 - 61 MB	CPU
CVT	TFLITE	float	Qualcomm® QCS8450	330.218 ms	0 - 2643 MB	NPU
CVT	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	280.469 ms	0 - 34 MB	NPU
CVT	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	179.294 ms	1 - 2095 MB	NPU
CVT	TFLITE	float	Qualcomm® SA8295P	315.727 ms	1 - 2263 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.