DETR-ResNet101: 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-ResNet101 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-ResNet101 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-ResNet101 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.object_detection

Model Stats:

• Model checkpoint: ResNet101
• Input resolution: 480x480
• Number of parameters: 60.3M
• Model size (float): 230 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
DETR-ResNet101	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	11.672 ms	5 - 383 MB	NPU
DETR-ResNet101	ONNX	float	Snapdragon® X2 Elite	12.82 ms	115 - 115 MB	NPU
DETR-ResNet101	ONNX	float	Snapdragon® X Elite	25.435 ms	114 - 114 MB	NPU
DETR-ResNet101	ONNX	float	Snapdragon® 8 Gen 3 Mobile	18.996 ms	5 - 497 MB	NPU
DETR-ResNet101	ONNX	float	Qualcomm® QCS8550 (Proxy)	25.159 ms	0 - 130 MB	NPU
DETR-ResNet101	ONNX	float	Qualcomm® QCS9075	40.365 ms	5 - 12 MB	NPU
DETR-ResNet101	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	14.998 ms	0 - 318 MB	NPU
DETR-ResNet101	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	12.379 ms	5 - 360 MB	NPU
DETR-ResNet101	QNN_DLC	float	Snapdragon® X2 Elite	13.884 ms	5 - 5 MB	NPU
DETR-ResNet101	QNN_DLC	float	Snapdragon® X Elite	28.314 ms	5 - 5 MB	NPU
DETR-ResNet101	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	20.52 ms	3 - 463 MB	NPU
DETR-ResNet101	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	139.859 ms	2 - 338 MB	NPU
DETR-ResNet101	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	27.732 ms	5 - 7 MB	NPU
DETR-ResNet101	QNN_DLC	float	Qualcomm® SA8775P	42.301 ms	1 - 336 MB	NPU
DETR-ResNet101	QNN_DLC	float	Qualcomm® QCS9075	46.805 ms	5 - 11 MB	NPU
DETR-ResNet101	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	58.755 ms	2 - 362 MB	NPU
DETR-ResNet101	QNN_DLC	float	Qualcomm® SA7255P	139.859 ms	2 - 338 MB	NPU
DETR-ResNet101	QNN_DLC	float	Qualcomm® SA8295P	44.907 ms	2 - 250 MB	NPU
DETR-ResNet101	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	16.026 ms	0 - 338 MB	NPU
DETR-ResNet101	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	10.852 ms	0 - 388 MB	NPU
DETR-ResNet101	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	19.478 ms	0 - 489 MB	NPU
DETR-ResNet101	TFLITE	float	Qualcomm® QCS8275 (Proxy)	137.596 ms	0 - 369 MB	NPU
DETR-ResNet101	TFLITE	float	Qualcomm® QCS8550 (Proxy)	26.277 ms	0 - 3 MB	NPU
DETR-ResNet101	TFLITE	float	Qualcomm® SA8775P	41.001 ms	0 - 367 MB	NPU
DETR-ResNet101	TFLITE	float	Qualcomm® QCS9075	40.67 ms	0 - 125 MB	NPU
DETR-ResNet101	TFLITE	float	Qualcomm® QCS8450 (Proxy)	54.861 ms	0 - 385 MB	NPU
DETR-ResNet101	TFLITE	float	Qualcomm® SA7255P	137.596 ms	0 - 369 MB	NPU
DETR-ResNet101	TFLITE	float	Qualcomm® SA8295P	42.969 ms	0 - 274 MB	NPU
DETR-ResNet101	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	15.188 ms	0 - 371 MB	NPU

License

• The license for the original implementation of DETR-ResNet101 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.