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 1.27.1	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 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:

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

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
decoder	ONNX	float	Snapdragon® X2 Elite	2.627 ms	4 - 4 MB	NPU
decoder	ONNX	float	Snapdragon® X Elite	6.162 ms	11 - 11 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	4.044 ms	0 - 269 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	14.683 ms	5 - 234 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	6.271 ms	4 - 8 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	5.908 ms	0 - 15 MB	NPU
decoder	ONNX	float	Qualcomm® QCS8450	14.683 ms	5 - 234 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	6.872 ms	4 - 7 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	6.162 ms	11 - 11 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	3.19 ms	0 - 234 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite Mobile	3.19 ms	0 - 234 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	2.545 ms	1 - 215 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X2 Elite	2.98 ms	4 - 4 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X Elite	5.839 ms	4 - 4 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	3.795 ms	4 - 224 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	11.736 ms	0 - 219 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	5.767 ms	4 - 11 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	5.435 ms	4 - 6 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8775P	6.389 ms	0 - 196 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8650P	6.389 ms	0 - 196 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8255P	6.389 ms	0 - 196 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8450	11.736 ms	0 - 219 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	6.172 ms	4 - 10 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	5.839 ms	4 - 4 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	2.857 ms	0 - 221 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8295P	7.816 ms	0 - 195 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Mobile	2.857 ms	0 - 221 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	2.559 ms	3 - 196 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	3.457 ms	0 - 213 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	8.33 ms	0 - 215 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	5.344 ms	0 - 19 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	5.002 ms	0 - 8 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8775P	5.89 ms	0 - 188 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8650P	5.89 ms	0 - 188 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8255P	5.89 ms	0 - 188 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8450	8.33 ms	0 - 215 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	5.676 ms	0 - 19 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	2.614 ms	0 - 189 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8295P	7.354 ms	0 - 190 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Mobile	2.614 ms	0 - 189 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	2.193 ms	0 - 186 MB	NPU
encoder	ONNX	float	Snapdragon® X2 Elite	59.351 ms	12 - 12 MB	NPU
encoder	ONNX	float	Snapdragon® X Elite	117.235 ms	13 - 13 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	83.755 ms	16 - 1137 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	324.771 ms	20 - 1190 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	112.124 ms	12 - 16 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	116.13 ms	12 - 30 MB	NPU
encoder	ONNX	float	Qualcomm® QCS8450	324.771 ms	20 - 1190 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	122.096 ms	12 - 15 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	117.235 ms	13 - 13 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	63.757 ms	8 - 927 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Mobile	63.757 ms	8 - 927 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	52.278 ms	8 - 1025 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X2 Elite	46.017 ms	12 - 12 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X Elite	102.875 ms	12 - 12 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	66.104 ms	12 - 1330 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	327.242 ms	13 - 1343 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	92.342 ms	12 - 30 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	99.328 ms	12 - 15 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8775P	100.287 ms	1 - 1549 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8650P	100.287 ms	1 - 1549 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8255P	100.287 ms	1 - 1549 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8450	327.242 ms	13 - 1343 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	101.525 ms	12 - 30 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	102.875 ms	12 - 12 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	49.197 ms	0 - 1116 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8295P	324.83 ms	0 - 1556 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Mobile	49.197 ms	0 - 1116 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	40.253 ms	12 - 1260 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	67.233 ms	3 - 3190 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	311.685 ms	0 - 1359 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	86.723 ms	0 - 47 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	90.42 ms	4 - 8 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8775P	100.639 ms	0 - 1665 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8650P	100.639 ms	0 - 1665 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8255P	100.639 ms	0 - 1665 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8450	311.685 ms	0 - 1359 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	101.725 ms	0 - 45 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	48.011 ms	2 - 1632 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8295P	258.452 ms	0 - 1710 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite Mobile	48.011 ms	2 - 1632 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	40.25 ms	4 - 1796 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.