MobileSam: Optimized for Qualcomm Devices

Transformer based encoder-decoder where prompts specify what to segment in an image thereby allowing segmentation without the need for additional training. The image encoder generates embeddings and the lightweight decoder operates on the embeddings for point and mask based image segmentation.

This is based on the implementation of MobileSam 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 1.25.0	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 MobileSam 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 MobileSam on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.semantic_segmentation

Model Stats:

• Model checkpoint: vit_t
• Input resolution: 720p (720x1280)
• Number of parameters (sam_encoder): 6.95M
• Model size (sam_encoder) (float): 26.6 MB
• Number of parameters (sam_decoder): 6.16M
• Model size (sam_decoder) (float): 23.7 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
decoder	ONNX	float	Snapdragon® X2 Elite	2.638 ms	4 - 4 MB	NPU
decoder	ONNX	float	Snapdragon® X Elite	6.168 ms	11 - 11 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	4.144 ms	5 - 245 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	14.332 ms	5 - 233 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	5.913 ms	0 - 14 MB	NPU
decoder	ONNX	float	Qualcomm® QCS8450	14.332 ms	5 - 233 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	6.815 ms	4 - 7 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	2.536 ms	2 - 213 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite Mobile	3.184 ms	1 - 235 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	3.184 ms	1 - 235 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	6.168 ms	11 - 11 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X2 Elite	2.964 ms	4 - 4 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X Elite	5.704 ms	4 - 4 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	3.734 ms	4 - 225 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	11.619 ms	0 - 252 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8275	12.566 ms	0 - 193 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	5.279 ms	4 - 109 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8775P	6.257 ms	0 - 194 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8650P	6.257 ms	0 - 194 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8255P	6.257 ms	0 - 194 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8450	11.619 ms	0 - 252 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	6.062 ms	4 - 10 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	2.539 ms	4 - 196 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA7255P	12.566 ms	0 - 193 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Mobile	2.832 ms	0 - 224 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8295P	7.717 ms	0 - 195 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	2.832 ms	0 - 224 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	5.704 ms	4 - 4 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	3.405 ms	0 - 212 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	7.99 ms	0 - 218 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8275	11.867 ms	0 - 186 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	4.945 ms	0 - 7 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8775P	5.929 ms	0 - 187 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8650P	5.929 ms	0 - 187 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8255P	5.929 ms	0 - 187 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8450	7.99 ms	0 - 218 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	5.716 ms	0 - 19 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	2.16 ms	0 - 185 MB	NPU
decoder	TFLITE	float	Qualcomm® SA7255P	11.867 ms	0 - 186 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Mobile	2.588 ms	0 - 183 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8295P	7.25 ms	0 - 183 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	2.588 ms	0 - 183 MB	NPU
encoder	ONNX	float	Snapdragon® X2 Elite	91.725 ms	12 - 12 MB	NPU
encoder	ONNX	float	Snapdragon® X Elite	166.555 ms	14 - 14 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	125.57 ms	12 - 1102 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	411.144 ms	13 - 1119 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	177.059 ms	12 - 16 MB	NPU
encoder	ONNX	float	Qualcomm® QCS8450	411.144 ms	13 - 1119 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	178.482 ms	12 - 15 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	81.762 ms	8 - 1033 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Mobile	98.908 ms	8 - 919 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	98.908 ms	8 - 919 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	166.555 ms	14 - 14 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X2 Elite	45.919 ms	12 - 12 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X Elite	102.883 ms	12 - 12 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	66.157 ms	0 - 1317 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	327.581 ms	12 - 1342 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8275	207.441 ms	1 - 1126 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	100.844 ms	12 - 15 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8775P	100.324 ms	0 - 1549 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8650P	100.324 ms	0 - 1549 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8255P	100.324 ms	0 - 1549 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8450	327.581 ms	12 - 1342 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	101.248 ms	12 - 30 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	40.255 ms	0 - 1248 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA7255P	207.441 ms	1 - 1126 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Mobile	49.634 ms	0 - 1119 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8295P	324.885 ms	1 - 1557 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	49.634 ms	0 - 1119 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	102.883 ms	12 - 12 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	172.777 ms	33 - 2313 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	518.348 ms	0 - 1941 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8275	358.191 ms	33 - 1120 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	223.083 ms	33 - 37 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8775P	242.001 ms	34 - 1114 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8650P	242.001 ms	34 - 1114 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8255P	242.001 ms	34 - 1114 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8450	518.348 ms	0 - 1941 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	246.161 ms	32 - 109 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	96.851 ms	32 - 1095 MB	NPU
encoder	TFLITE	float	Qualcomm® SA7255P	358.191 ms	33 - 1120 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite Mobile	129.438 ms	32 - 1113 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8295P	371.6 ms	34 - 1161 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	129.438 ms	32 - 1113 MB	NPU

License

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

References

• Segment Anything
• 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.