DnCNN: Optimized for Qualcomm Devices

DnCNN is a 17-layer denoising convolutional neural network that uses residual learning to remove Gaussian noise (sigma=25) from grayscale images. The network predicts the noise residual and subtracts it from the input to produce a clean image.

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

Model Details

Model Type: Model_use_case.image_editing

Model Stats:

• Input resolution: 256x256
• Model checkpoint: dncnn_25
• Model size (float): 2.12 MB
• Model size (w8a8): 581 KB
• Number of parameters: 555K

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
DnCNN	ONNX	float	Snapdragon® X2 Elite	4.046 ms	1 - 1 MB	NPU
DnCNN	ONNX	float	Snapdragon® X Elite	7.16 ms	0 - 0 MB	NPU
DnCNN	ONNX	float	Snapdragon® 8 Gen 3 Mobile	5.158 ms	1 - 180 MB	NPU
DnCNN	ONNX	float	Snapdragon® 8 Gen 1 Mobile	13.764 ms	1 - 179 MB	NPU
DnCNN	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	16.04 ms	1 - 4 MB	NPU
DnCNN	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	6.891 ms	1 - 3 MB	NPU
DnCNN	ONNX	float	Qualcomm® QCS8450	13.764 ms	1 - 179 MB	NPU
DnCNN	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	14.495 ms	1 - 4 MB	NPU
DnCNN	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	7.16 ms	0 - 0 MB	NPU
DnCNN	ONNX	float	Qualcomm® Dragonwing™ Q-8750	4.124 ms	0 - 144 MB	NPU
DnCNN	ONNX	float	Snapdragon® 8 Elite Mobile	4.124 ms	0 - 144 MB	NPU
DnCNN	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	3.212 ms	0 - 144 MB	NPU
DnCNN	QNN_DLC	float	Snapdragon® X2 Elite	4.153 ms	0 - 0 MB	NPU
DnCNN	QNN_DLC	float	Snapdragon® X Elite	7.233 ms	0 - 0 MB	NPU
DnCNN	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	5.047 ms	0 - 176 MB	NPU
DnCNN	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	13.519 ms	0 - 178 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	15.832 ms	0 - 3 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	55.996 ms	0 - 141 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	6.676 ms	0 - 2 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® SA8775P	13.879 ms	0 - 143 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® SA8650P	13.879 ms	0 - 143 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® SA8255P	13.879 ms	0 - 143 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® QCS8450	13.519 ms	0 - 178 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	14.14 ms	2 - 4 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	7.233 ms	0 - 0 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	4.035 ms	0 - 144 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® SA7255P	55.996 ms	0 - 141 MB	NPU
DnCNN	QNN_DLC	float	Qualcomm® SA8295P	15.289 ms	0 - 139 MB	NPU
DnCNN	QNN_DLC	float	Snapdragon® 8 Elite Mobile	4.035 ms	0 - 144 MB	NPU
DnCNN	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	3.027 ms	0 - 148 MB	NPU
DnCNN	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	5.197 ms	0 - 179 MB	NPU
DnCNN	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	13.906 ms	0 - 178 MB	NPU
DnCNN	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	15.918 ms	0 - 5 MB	NPU
DnCNN	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	56.399 ms	0 - 142 MB	NPU
DnCNN	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	6.929 ms	0 - 11 MB	NPU
DnCNN	TFLITE	float	Qualcomm® SA8775P	14.161 ms	0 - 145 MB	NPU
DnCNN	TFLITE	float	Qualcomm® SA8650P	14.161 ms	0 - 145 MB	NPU
DnCNN	TFLITE	float	Qualcomm® SA8255P	14.161 ms	0 - 145 MB	NPU
DnCNN	TFLITE	float	Qualcomm® QCS8450	13.906 ms	0 - 178 MB	NPU
DnCNN	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	14.024 ms	0 - 4 MB	NPU
DnCNN	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	4.13 ms	0 - 142 MB	NPU
DnCNN	TFLITE	float	Qualcomm® SA7255P	56.399 ms	0 - 142 MB	NPU
DnCNN	TFLITE	float	Qualcomm® SA8295P	15.603 ms	0 - 141 MB	NPU
DnCNN	TFLITE	float	Snapdragon® 8 Elite Mobile	4.13 ms	0 - 142 MB	NPU
DnCNN	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	3.137 ms	0 - 147 MB	NPU

License

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

References

• Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
• 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.