DenseNet-121: Optimized for Qualcomm Devices

Densenet is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.

This is based on the implementation of DenseNet-121 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 2.37, ONNX Runtime 1.23.0	Download
QNN_DLC	float	Universal	QAIRT 2.42	Download
QNN_DLC	w8a16	Universal	QAIRT 2.42	Download
QNN_DLC	w8a8	Universal	QAIRT 2.42	Download
TFLITE	float	Universal	QAIRT 2.42, TFLite 2.17.0	Download
TFLITE	w8a8	Universal	QAIRT 2.42, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit DenseNet-121 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 DenseNet-121 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.image_classification

Model Stats:

• Model checkpoint: Imagenet
• Input resolution: 224x224
• Number of parameters: 7.99M
• Model size (float): 30.5 MB
• Model size (w8a16): 8.72 MB
• Model size (w8a8): 8.30 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
DenseNet-121	ONNX	float	Snapdragon® X Elite	1.741 ms	15 - 15 MB	NPU
DenseNet-121	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.298 ms	0 - 155 MB	NPU
DenseNet-121	ONNX	float	Qualcomm® QCS8550 (Proxy)	1.788 ms	0 - 37 MB	NPU
DenseNet-121	ONNX	float	Qualcomm® QCS9075	2.872 ms	1 - 3 MB	NPU
DenseNet-121	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	0.957 ms	0 - 110 MB	NPU
DenseNet-121	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.83 ms	0 - 109 MB	NPU
DenseNet-121	QNN_DLC	float	Snapdragon® X Elite	2.077 ms	1 - 1 MB	NPU
DenseNet-121	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.24 ms	0 - 79 MB	NPU
DenseNet-121	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	7.958 ms	1 - 39 MB	NPU
DenseNet-121	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	1.835 ms	1 - 2 MB	NPU
DenseNet-121	QNN_DLC	float	Qualcomm® SA8775P	2.786 ms	1 - 43 MB	NPU
DenseNet-121	QNN_DLC	float	Qualcomm® QCS9075	2.754 ms	1 - 3 MB	NPU
DenseNet-121	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	3.28 ms	0 - 75 MB	NPU
DenseNet-121	QNN_DLC	float	Qualcomm® SA7255P	7.958 ms	1 - 39 MB	NPU
DenseNet-121	QNN_DLC	float	Qualcomm® SA8295P	3.147 ms	0 - 36 MB	NPU
DenseNet-121	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	0.884 ms	0 - 43 MB	NPU
DenseNet-121	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.71 ms	1 - 44 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Snapdragon® X Elite	3.433 ms	0 - 0 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	2.134 ms	0 - 88 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	6.217 ms	0 - 62 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	3.126 ms	0 - 2 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Qualcomm® SA8775P	3.489 ms	0 - 65 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Qualcomm® QCS9075	3.39 ms	0 - 2 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Qualcomm® QCS8450 (Proxy)	5.014 ms	0 - 86 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Qualcomm® SA7255P	6.217 ms	0 - 62 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Qualcomm® SA8295P	5.535 ms	0 - 61 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	1.755 ms	0 - 59 MB	NPU
DenseNet-121	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	1.218 ms	0 - 72 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Snapdragon® X Elite	2.663 ms	0 - 0 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	1.717 ms	0 - 81 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	4.99 ms	0 - 58 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	2.394 ms	0 - 2 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Qualcomm® SA8775P	2.736 ms	0 - 60 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Qualcomm® QCS9075	2.558 ms	2 - 4 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	2.942 ms	0 - 77 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Qualcomm® SA7255P	4.99 ms	0 - 58 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Qualcomm® SA8295P	3.519 ms	0 - 56 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	1.038 ms	0 - 59 MB	NPU
DenseNet-121	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.925 ms	0 - 64 MB	NPU
DenseNet-121	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1.244 ms	36 - 132 MB	NPU
DenseNet-121	TFLITE	float	Qualcomm® QCS8275 (Proxy)	7.977 ms	0 - 56 MB	NPU
DenseNet-121	TFLITE	float	Qualcomm® QCS8550 (Proxy)	1.807 ms	0 - 2 MB	NPU
DenseNet-121	TFLITE	float	Qualcomm® SA8775P	2.762 ms	0 - 58 MB	NPU
DenseNet-121	TFLITE	float	Qualcomm® QCS9075	2.805 ms	0 - 18 MB	NPU
DenseNet-121	TFLITE	float	Qualcomm® QCS8450 (Proxy)	3.325 ms	0 - 97 MB	NPU
DenseNet-121	TFLITE	float	Qualcomm® SA7255P	7.977 ms	0 - 56 MB	NPU
DenseNet-121	TFLITE	float	Qualcomm® SA8295P	3.163 ms	0 - 52 MB	NPU
DenseNet-121	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	0.887 ms	0 - 53 MB	NPU
DenseNet-121	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.712 ms	0 - 59 MB	NPU
DenseNet-121	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	1.611 ms	0 - 86 MB	NPU
DenseNet-121	TFLITE	w8a8	Qualcomm® QCS6490	63.536 ms	6 - 34 MB	NPU
DenseNet-121	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	4.674 ms	0 - 60 MB	NPU
DenseNet-121	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	2.228 ms	0 - 3 MB	NPU
DenseNet-121	TFLITE	w8a8	Qualcomm® SA8775P	2.621 ms	0 - 62 MB	NPU
DenseNet-121	TFLITE	w8a8	Qualcomm® QCS9075	2.419 ms	0 - 11 MB	NPU
DenseNet-121	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	2.635 ms	0 - 80 MB	NPU
DenseNet-121	TFLITE	w8a8	Qualcomm® SA7255P	4.674 ms	0 - 60 MB	NPU
DenseNet-121	TFLITE	w8a8	Qualcomm® SA8295P	3.352 ms	0 - 57 MB	NPU
DenseNet-121	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	1.008 ms	0 - 63 MB	NPU
DenseNet-121	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	27.248 ms	6 - 171 MB	NPU
DenseNet-121	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.875 ms	0 - 66 MB	NPU

License

• The license for the original implementation of DenseNet-121 can be found here.

References

• Densely Connected Convolutional Networks
• 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.