SESR-M5: Optimized for Qualcomm Devices

SESR M5 performs efficient on-device upscaling of images.

This is based on the implementation of SESR-M5 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 1.24.1	Download
ONNX	w8a8	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
QNN_DLC	w8a8	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.17.0	Download
TFLITE	w8a8	Universal	QAIRT 2.43, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit SESR-M5 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 SESR-M5 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.super_resolution

Model Stats:

• Model checkpoint: sesr_m5_3x_checkpoint
• Input resolution: 128x128
• Number of parameters: 343K
• Model size (float): 1.32 MB
• Model size (w8a8): 395 KB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
SESR-M5	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.969 ms	0 - 25 MB	NPU
SESR-M5	ONNX	float	Snapdragon® X2 Elite	1.144 ms	6 - 6 MB	NPU
SESR-M5	ONNX	float	Snapdragon® X Elite	2.203 ms	8 - 8 MB	NPU
SESR-M5	ONNX	float	Qualcomm® QCS8550 (Proxy)	2.0 ms	0 - 2 MB	NPU
SESR-M5	ONNX	float	Qualcomm® QCS9075	3.236 ms	6 - 9 MB	NPU
SESR-M5	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	1.206 ms	0 - 21 MB	NPU
SESR-M5	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.378 ms	0 - 26 MB	NPU
SESR-M5	ONNX	w8a8	Snapdragon® X2 Elite	0.407 ms	1 - 1 MB	NPU
SESR-M5	ONNX	w8a8	Snapdragon® X Elite	0.929 ms	3 - 3 MB	NPU
SESR-M5	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	0.533 ms	0 - 33 MB	NPU
SESR-M5	ONNX	w8a8	Qualcomm® QCS6490	42.395 ms	38 - 41 MB	CPU
SESR-M5	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	0.785 ms	0 - 24 MB	NPU
SESR-M5	ONNX	w8a8	Qualcomm® QCS9075	0.981 ms	0 - 3 MB	NPU
SESR-M5	ONNX	w8a8	Qualcomm® QCM6690	31.12 ms	38 - 44 MB	CPU
SESR-M5	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.427 ms	0 - 27 MB	NPU
SESR-M5	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	23.759 ms	38 - 45 MB	CPU
SESR-M5	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.864 ms	0 - 27 MB	NPU
SESR-M5	QNN_DLC	float	Snapdragon® X2 Elite	1.19 ms	0 - 0 MB	NPU
SESR-M5	QNN_DLC	float	Snapdragon® X Elite	2.064 ms	0 - 0 MB	NPU
SESR-M5	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.275 ms	0 - 35 MB	NPU
SESR-M5	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	10.546 ms	0 - 23 MB	NPU
SESR-M5	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	1.882 ms	0 - 26 MB	NPU
SESR-M5	QNN_DLC	float	Qualcomm® SA8775P	3.015 ms	0 - 25 MB	NPU
SESR-M5	QNN_DLC	float	Qualcomm® QCS9075	3.046 ms	0 - 5 MB	NPU
SESR-M5	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	2.796 ms	0 - 37 MB	NPU
SESR-M5	QNN_DLC	float	Qualcomm® SA7255P	10.546 ms	0 - 23 MB	NPU
SESR-M5	QNN_DLC	float	Qualcomm® SA8295P	3.469 ms	0 - 21 MB	NPU
SESR-M5	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	1.04 ms	0 - 28 MB	NPU
SESR-M5	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.301 ms	0 - 24 MB	NPU
SESR-M5	QNN_DLC	w8a8	Snapdragon® X2 Elite	0.448 ms	0 - 0 MB	NPU
SESR-M5	QNN_DLC	w8a8	Snapdragon® X Elite	0.773 ms	0 - 0 MB	NPU
SESR-M5	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	0.443 ms	0 - 31 MB	NPU
SESR-M5	QNN_DLC	w8a8	Qualcomm® QCS6490	2.736 ms	0 - 2 MB	NPU
SESR-M5	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	1.946 ms	0 - 22 MB	NPU
SESR-M5	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	0.623 ms	0 - 52 MB	NPU
SESR-M5	QNN_DLC	w8a8	Qualcomm® SA8775P	0.852 ms	0 - 23 MB	NPU
SESR-M5	QNN_DLC	w8a8	Qualcomm® QCS9075	0.885 ms	0 - 2 MB	NPU
SESR-M5	QNN_DLC	w8a8	Qualcomm® QCM6690	6.821 ms	0 - 135 MB	NPU
SESR-M5	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	0.941 ms	0 - 32 MB	NPU
SESR-M5	QNN_DLC	w8a8	Qualcomm® SA7255P	1.946 ms	0 - 22 MB	NPU
SESR-M5	QNN_DLC	w8a8	Qualcomm® SA8295P	1.393 ms	0 - 20 MB	NPU
SESR-M5	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.38 ms	0 - 21 MB	NPU
SESR-M5	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	0.84 ms	0 - 21 MB	NPU
SESR-M5	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.866 ms	0 - 28 MB	NPU
SESR-M5	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1.281 ms	0 - 35 MB	NPU
SESR-M5	TFLITE	float	Qualcomm® QCS8275 (Proxy)	10.563 ms	3 - 27 MB	NPU
SESR-M5	TFLITE	float	Qualcomm® QCS8550 (Proxy)	1.888 ms	0 - 2 MB	NPU
SESR-M5	TFLITE	float	Qualcomm® SA8775P	3.078 ms	0 - 26 MB	NPU
SESR-M5	TFLITE	float	Qualcomm® QCS9075	3.063 ms	3 - 9 MB	NPU
SESR-M5	TFLITE	float	Qualcomm® QCS8450 (Proxy)	2.787 ms	0 - 37 MB	NPU
SESR-M5	TFLITE	float	Qualcomm® SA7255P	10.563 ms	3 - 27 MB	NPU
SESR-M5	TFLITE	float	Qualcomm® SA8295P	3.498 ms	0 - 22 MB	NPU
SESR-M5	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	1.038 ms	0 - 24 MB	NPU
SESR-M5	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	0.46 ms	0 - 30 MB	NPU
SESR-M5	TFLITE	w8a8	Qualcomm® QCS6490	2.704 ms	0 - 3 MB	NPU
SESR-M5	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	0.638 ms	0 - 1 MB	NPU
SESR-M5	TFLITE	w8a8	Qualcomm® SA8775P	0.887 ms	0 - 24 MB	NPU
SESR-M5	TFLITE	w8a8	Qualcomm® QCS9075	0.9 ms	0 - 3 MB	NPU
SESR-M5	TFLITE	w8a8	Qualcomm® QCM6690	6.837 ms	1 - 135 MB	NPU
SESR-M5	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	0.947 ms	0 - 33 MB	NPU
SESR-M5	TFLITE	w8a8	Qualcomm® SA8295P	1.396 ms	0 - 20 MB	NPU
SESR-M5	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.377 ms	0 - 21 MB	NPU

License

• The license for the original implementation of SESR-M5 can be found here.

References

• Collapsible Linear Blocks for Super-Efficient Super Resolution
• 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.