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 2.45, ONNX Runtime 1.27.1	Download
ONNX	w8a16_mixed_fp16	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 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® X2 Elite	8.252 ms	5 - 5 MB	NPU
DETR-ResNet50	ONNX	float	Snapdragon® X Elite	17.808 ms	77 - 77 MB	NPU
DETR-ResNet50	ONNX	float	Snapdragon® 8 Gen 3 Mobile	13.391 ms	2 - 372 MB	NPU
DETR-ResNet50	ONNX	float	Snapdragon® 8 Gen 1 Mobile	36.011 ms	3 - 319 MB	NPU
DETR-ResNet50	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	17.87 ms	0 - 95 MB	NPU
DETR-ResNet50	ONNX	float	Qualcomm® QCS8450	36.011 ms	3 - 319 MB	NPU
DETR-ResNet50	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	27.053 ms	4 - 12 MB	NPU
DETR-ResNet50	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	17.808 ms	77 - 77 MB	NPU
DETR-ResNet50	ONNX	float	Qualcomm® Dragonwing™ Q-8750	9.971 ms	2 - 271 MB	NPU
DETR-ResNet50	ONNX	float	Snapdragon® 8 Elite Mobile	9.971 ms	2 - 271 MB	NPU
DETR-ResNet50	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	7.428 ms	2 - 316 MB	NPU
DETR-ResNet50	ONNX	w8a16_mixed_fp16	Snapdragon® 8 Gen 3 Mobile	47.583 ms	13 - 473 MB	NPU
DETR-ResNet50	ONNX	w8a16_mixed_fp16	Snapdragon® 8 Gen 1 Mobile	97.068 ms	7 - 466 MB	NPU
DETR-ResNet50	ONNX	w8a16_mixed_fp16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	62.487 ms	0 - 58 MB	NPU
DETR-ResNet50	ONNX	w8a16_mixed_fp16	Qualcomm® QCS8450	97.068 ms	7 - 466 MB	NPU
DETR-ResNet50	ONNX	w8a16_mixed_fp16	Qualcomm® Dragonwing™ IQ-9075	70.286 ms	16 - 19 MB	NPU
DETR-ResNet50	ONNX	w8a16_mixed_fp16	Qualcomm® Dragonwing™ Q-8750	39.704 ms	15 - 400 MB	NPU
DETR-ResNet50	ONNX	w8a16_mixed_fp16	Snapdragon® 8 Elite Mobile	39.704 ms	15 - 400 MB	NPU
DETR-ResNet50	ONNX	w8a16_mixed_fp16	Snapdragon® 8 Elite Gen 5 Mobile	32.585 ms	15 - 453 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® X2 Elite	9.012 ms	5 - 5 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® X Elite	19.521 ms	5 - 5 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	13.649 ms	5 - 355 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	41.728 ms	5 - 305 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8275	85.978 ms	1 - 268 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	19.268 ms	5 - 312 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® SA8775P	26.859 ms	0 - 299 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® SA8650P	26.859 ms	0 - 299 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® SA8255P	26.859 ms	0 - 299 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® QCS8450	41.728 ms	5 - 305 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	27.255 ms	5 - 11 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	19.521 ms	5 - 5 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	10.445 ms	5 - 280 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® SA7255P	85.978 ms	1 - 268 MB	NPU
DETR-ResNet50	QNN_DLC	float	Qualcomm® SA8295P	31.186 ms	1 - 225 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® 8 Elite Mobile	10.445 ms	5 - 280 MB	NPU
DETR-ResNet50	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	7.909 ms	5 - 291 MB	NPU
DETR-ResNet50	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	13.627 ms	0 - 385 MB	NPU
DETR-ResNet50	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	42.411 ms	0 - 328 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® Dragonwing™ QCS8275	85.989 ms	0 - 293 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	19.189 ms	0 - 10 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® SA8775P	26.976 ms	0 - 324 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® SA8650P	26.976 ms	0 - 324 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® SA8255P	26.976 ms	0 - 324 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® QCS8450	42.411 ms	0 - 328 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	27.211 ms	0 - 88 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	10.485 ms	0 - 307 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® SA7255P	85.989 ms	0 - 293 MB	NPU
DETR-ResNet50	TFLITE	float	Qualcomm® SA8295P	31.067 ms	0 - 247 MB	NPU
DETR-ResNet50	TFLITE	float	Snapdragon® 8 Elite Mobile	10.485 ms	0 - 307 MB	NPU
DETR-ResNet50	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	8.046 ms	0 - 312 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.