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 2.37, ONNX Runtime 1.23.0	Download
QNN_DLC	float	Universal	QAIRT 2.42	Download
TFLITE	float	Universal	QAIRT 2.42, TFLite 2.17.0	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 (SAMEncoder): 6.95M
• Model size (SAMEncoder) (float): 26.6 MB
• Number of parameters (SAMDecoder): 6.16M
• Model size (SAMDecoder) (float): 23.7 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
MobileSAMDecoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	5.734 ms	5 - 218 MB	NPU
MobileSAMDecoder	ONNX	float	Qualcomm® QCS8550 (Proxy)	8.389 ms	4 - 230 MB	NPU
MobileSAMDecoder	ONNX	float	Qualcomm® QCS9075	10.04 ms	4 - 7 MB	NPU
MobileSAMDecoder	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	4.263 ms	2 - 180 MB	NPU
MobileSAMDecoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	3.481 ms	0 - 186 MB	NPU
MobileSAMDecoder	QNN_DLC	float	Snapdragon® X Elite	5.137 ms	4 - 4 MB	NPU
MobileSAMDecoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	3.319 ms	4 - 230 MB	NPU
MobileSAMDecoder	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	11.595 ms	2 - 199 MB	NPU
MobileSAMDecoder	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	4.692 ms	4 - 6 MB	NPU
MobileSAMDecoder	QNN_DLC	float	Qualcomm® SA8775P	5.737 ms	1 - 200 MB	NPU
MobileSAMDecoder	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	8.499 ms	0 - 222 MB	NPU
MobileSAMDecoder	QNN_DLC	float	Qualcomm® SA7255P	11.595 ms	2 - 199 MB	NPU
MobileSAMDecoder	QNN_DLC	float	Qualcomm® SA8295P	7.157 ms	0 - 193 MB	NPU
MobileSAMDecoder	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	2.538 ms	0 - 198 MB	NPU
MobileSAMDecoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	2.182 ms	4 - 203 MB	NPU
MobileSAMDecoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	3.854 ms	0 - 237 MB	NPU
MobileSAMDecoder	TFLITE	float	Qualcomm® QCS8275 (Proxy)	12.771 ms	0 - 208 MB	NPU
MobileSAMDecoder	TFLITE	float	Qualcomm® QCS8550 (Proxy)	5.561 ms	0 - 18 MB	NPU
MobileSAMDecoder	TFLITE	float	Qualcomm® SA8775P	6.547 ms	0 - 209 MB	NPU
MobileSAMDecoder	TFLITE	float	Qualcomm® QCS8450 (Proxy)	9.544 ms	0 - 225 MB	NPU
MobileSAMDecoder	TFLITE	float	Qualcomm® SA7255P	12.771 ms	0 - 208 MB	NPU
MobileSAMDecoder	TFLITE	float	Qualcomm® SA8295P	8.152 ms	0 - 205 MB	NPU
MobileSAMDecoder	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	2.935 ms	0 - 209 MB	NPU
MobileSAMEncoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	254.846 ms	128 - 866 MB	NPU
MobileSAMEncoder	ONNX	float	Qualcomm® QCS8550 (Proxy)	341.905 ms	114 - 116 MB	NPU
MobileSAMEncoder	ONNX	float	Qualcomm® QCS9075	414.464 ms	68 - 71 MB	NPU
MobileSAMEncoder	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	185.472 ms	117 - 762 MB	NPU
MobileSAMEncoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	152.087 ms	130 - 692 MB	NPU
MobileSAMEncoder	QNN_DLC	float	Snapdragon® X Elite	99.194 ms	12 - 12 MB	NPU
MobileSAMEncoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	65.564 ms	11 - 1629 MB	NPU
MobileSAMEncoder	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	209.778 ms	0 - 1209 MB	NPU
MobileSAMEncoder	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	96.271 ms	12 - 16 MB	NPU
MobileSAMEncoder	QNN_DLC	float	Qualcomm® SA8775P	102.588 ms	0 - 1614 MB	NPU
MobileSAMEncoder	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	467.917 ms	12 - 1732 MB	NPU
MobileSAMEncoder	QNN_DLC	float	Qualcomm® SA7255P	209.778 ms	0 - 1209 MB	NPU
MobileSAMEncoder	QNN_DLC	float	Qualcomm® SA8295P	474.029 ms	0 - 1248 MB	NPU
MobileSAMEncoder	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	49.974 ms	12 - 1214 MB	NPU
MobileSAMEncoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	44.853 ms	11 - 1322 MB	NPU
MobileSAMEncoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	127.122 ms	4 - 1521 MB	NPU
MobileSAMEncoder	TFLITE	float	Qualcomm® QCS8275 (Proxy)	292.444 ms	4 - 1275 MB	NPU
MobileSAMEncoder	TFLITE	float	Qualcomm® QCS8550 (Proxy)	176.108 ms	0 - 3 MB	NPU
MobileSAMEncoder	TFLITE	float	Qualcomm® SA8775P	180.445 ms	4 - 1275 MB	NPU
MobileSAMEncoder	TFLITE	float	Qualcomm® QCS8450 (Proxy)	588.02 ms	0 - 3179 MB	NPU
MobileSAMEncoder	TFLITE	float	Qualcomm® SA7255P	292.444 ms	4 - 1275 MB	NPU
MobileSAMEncoder	TFLITE	float	Qualcomm® SA8295P	504.031 ms	4 - 1899 MB	NPU
MobileSAMEncoder	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	96.546 ms	3 - 1271 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.