NASNet: Optimized for Qualcomm Devices

NASNet is a CNN-based architecture discovered via Neural Architecture Search (NAS) that can classify images from the Imagenet dataset.

This is based on the implementation of NASNet 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 NASNet 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 NASNet on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.image_classification

Model Stats:

• Activations (M): 19.4
• GMACs: 5.9
• Input resolution: 224x224
• Model checkpoint: nasnetalarge.tf_in1k
• Model size (float): 338 MB
• Number of parameters: 88.7M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
NASNet	ONNX	float	Snapdragon® X2 Elite	9.67 ms	1 - 1 MB	NPU
NASNet	ONNX	float	Snapdragon® X Elite	18.957 ms	189 - 189 MB	NPU
NASNet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	14.205 ms	0 - 572 MB	NPU
NASNet	ONNX	float	Snapdragon® 8 Gen 1 Mobile	53.926 ms	2 - 520 MB	NPU
NASNet	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	30.173 ms	1 - 6 MB	NPU
NASNet	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	18.922 ms	0 - 193 MB	NPU
NASNet	ONNX	float	Qualcomm® QCS8450	53.926 ms	2 - 520 MB	NPU
NASNet	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	27.14 ms	1 - 5 MB	NPU
NASNet	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	18.957 ms	189 - 189 MB	NPU
NASNet	ONNX	float	Qualcomm® Dragonwing™ Q-8750	11.69 ms	1 - 406 MB	NPU
NASNet	ONNX	float	Snapdragon® 8 Elite Mobile	11.69 ms	1 - 406 MB	NPU
NASNet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	9.493 ms	1 - 418 MB	NPU
NASNet	QNN_DLC	float	Snapdragon® X2 Elite	8.835 ms	1 - 1 MB	NPU
NASNet	QNN_DLC	float	Snapdragon® X Elite	16.779 ms	1 - 1 MB	NPU
NASNet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	12.318 ms	0 - 513 MB	NPU
NASNet	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	48.508 ms	0 - 474 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	27.631 ms	1 - 5 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	16.25 ms	1 - 4 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® SA8775P	24.414 ms	1 - 355 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® SA8650P	24.414 ms	1 - 355 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® SA8255P	24.414 ms	1 - 355 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® QCS8450	48.508 ms	0 - 474 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	24.615 ms	1 - 4 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	16.779 ms	1 - 1 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	9.759 ms	1 - 350 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® SA7255P	89.205 ms	2 - 350 MB	NPU
NASNet	QNN_DLC	float	Qualcomm® SA8295P	38.601 ms	1 - 310 MB	NPU
NASNet	QNN_DLC	float	Snapdragon® 8 Elite Mobile	9.759 ms	1 - 350 MB	NPU
NASNet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	7.804 ms	0 - 367 MB	NPU
NASNet	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	12.615 ms	0 - 694 MB	NPU
NASNet	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	50.191 ms	0 - 652 MB	NPU
NASNet	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	27.851 ms	0 - 193 MB	NPU
NASNet	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	16.542 ms	4 - 201 MB	NPU
NASNet	TFLITE	float	Qualcomm® SA8775P	24.813 ms	0 - 537 MB	NPU
NASNet	TFLITE	float	Qualcomm® SA8650P	24.813 ms	0 - 537 MB	NPU
NASNet	TFLITE	float	Qualcomm® SA8255P	24.813 ms	0 - 537 MB	NPU
NASNet	TFLITE	float	Qualcomm® QCS8450	50.191 ms	0 - 652 MB	NPU
NASNet	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	24.558 ms	0 - 192 MB	NPU
NASNet	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	10.486 ms	0 - 534 MB	NPU
NASNet	TFLITE	float	Qualcomm® SA7255P	89.677 ms	0 - 538 MB	NPU
NASNet	TFLITE	float	Qualcomm® SA8295P	40.111 ms	0 - 486 MB	NPU
NASNet	TFLITE	float	Snapdragon® 8 Elite Mobile	10.486 ms	0 - 534 MB	NPU
NASNet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	8.248 ms	0 - 549 MB	NPU

License

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

References

• Learning Transferable Architectures for Scalable Image Recognition
• 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.