DETR-ResNet50: Optimized for Qualcomm Devices

DETR is a machine learning model that can detect objects (trained on COCO dataset).

This is based on the implementation of DETR-ResNet50 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 2.42, ONNX Runtime 1.24.3	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 DETR-ResNet50 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 DETR-ResNet50 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.object_detection

Model Stats:

• Model checkpoint: ResNet50
• Input resolution: 480x480
• Number of parameters: 41.4M
• Model size (float): 158 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
DETR-ResNet50	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	7.572 ms	3 - 323 MB	NPU
DETR-ResNet50	ONNX	float	Snapdragon® X2 Elite	8.498 ms	77 - 77 MB	NPU
DETR-ResNet50	ONNX	float	Snapdragon® X Elite	17.744 ms	77 - 77 MB	NPU
DETR-ResNet50	ONNX	float	Snapdragon® 8 Gen 3 Mobile	13.797 ms	1 - 399 MB	NPU
DETR-ResNet50	ONNX	float	Qualcomm® QCS8550 (Proxy)	17.641 ms	0 - 95 MB	NPU
DETR-ResNet50	ONNX	float	Qualcomm® QCS9075	28.732 ms	5 - 12 MB	NPU
DETR-ResNet50	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	10.668 ms	3 - 279 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	8.227 ms	4 - 302 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® X2 Elite	9.32 ms	5 - 5 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® X Elite	21.509 ms	5 - 5 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	15.677 ms	5 - 379 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	92.915 ms	1 - 285 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	21.065 ms	5 - 7 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® SA8775P	30.2 ms	1 - 284 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® QCS9075	32.265 ms	7 - 13 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	44.448 ms	2 - 321 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® SA7255P	92.915 ms	1 - 285 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® SA8295P	32.847 ms	1 - 239 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	11.272 ms	0 - 332 MB	NPU
DETR-ResNet50	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	7.859 ms	0 - 309 MB	NPU
DETR-ResNet50	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	13.622 ms	0 - 391 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® QCS8275 (Proxy)	85.499 ms	0 - 294 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® QCS8550 (Proxy)	18.909 ms	0 - 3 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® SA8775P	26.754 ms	0 - 330 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® QCS9075	29.713 ms	0 - 88 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® QCS8450 (Proxy)	41.5 ms	0 - 328 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® SA7255P	85.499 ms	0 - 294 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® SA8295P	31.008 ms	0 - 247 MB	NPU
DETR-ResNet50	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	10.35 ms	0 - 304 MB	NPU

License

• The license for the original implementation of DETR-ResNet50 can be found here.

References

• End-to-End Object Detection with Transformers
• 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.