FastSam-X: Optimized for Qualcomm Devices

The Fast Segment Anything Model (FastSAM) is a novel, real-time CNN-based solution for the Segment Anything task. This task is designed to segment any object within an image based on various possible user interaction prompts. The model performs competitively despite significantly reduced computation, making it a practical choice for a variety of vision tasks.

This is based on the implementation of FastSam-X 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 FastSam-X 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 FastSam-X on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.semantic_segmentation

Model Stats:

• Model checkpoint: fastsam-x.pt
• Inference latency: RealTime
• Input resolution: 640x640
• Number of parameters: 72.2M
• Model size (float): 276 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
FastSam-X	ONNX	float	Snapdragon® X Elite	46.486 ms	139 - 139 MB	NPU
FastSam-X	ONNX	float	Snapdragon® 8 Gen 3 Mobile	36.497 ms	4 - 267 MB	NPU
FastSam-X	ONNX	float	Qualcomm® QCS8550 (Proxy)	46.077 ms	11 - 14 MB	NPU
FastSam-X	ONNX	float	Qualcomm® QCS9075	73.748 ms	11 - 19 MB	NPU
FastSam-X	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	27.37 ms	12 - 187 MB	NPU
FastSam-X	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	18.362 ms	1 - 183 MB	NPU
FastSam-X	QNN_DLC	float	Snapdragon® X Elite	43.841 ms	5 - 5 MB	NPU
FastSam-X	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	32.661 ms	3 - 314 MB	NPU
FastSam-X	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	279.828 ms	2 - 223 MB	NPU
FastSam-X	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	43.11 ms	5 - 7 MB	NPU
FastSam-X	QNN_DLC	float	Qualcomm® SA8775P	68.478 ms	0 - 222 MB	NPU
FastSam-X	QNN_DLC	float	Qualcomm® QCS9075	70.434 ms	7 - 17 MB	NPU
FastSam-X	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	93.211 ms	2 - 392 MB	NPU
FastSam-X	QNN_DLC	float	Qualcomm® SA7255P	279.828 ms	2 - 223 MB	NPU
FastSam-X	QNN_DLC	float	Qualcomm® SA8295P	77.966 ms	0 - 296 MB	NPU
FastSam-X	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	25.29 ms	0 - 222 MB	NPU
FastSam-X	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	17.889 ms	5 - 242 MB	NPU
FastSam-X	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	32.534 ms	3 - 443 MB	NPU
FastSam-X	TFLITE	float	Qualcomm® QCS8275 (Proxy)	279.179 ms	4 - 269 MB	NPU
FastSam-X	TFLITE	float	Qualcomm® QCS8550 (Proxy)	42.096 ms	4 - 43 MB	NPU
FastSam-X	TFLITE	float	Qualcomm® SA8775P	68.042 ms	4 - 269 MB	NPU
FastSam-X	TFLITE	float	Qualcomm® QCS9075	70.216 ms	4 - 158 MB	NPU
FastSam-X	TFLITE	float	Qualcomm® QCS8450 (Proxy)	92.525 ms	5 - 525 MB	NPU
FastSam-X	TFLITE	float	Qualcomm® SA7255P	279.179 ms	4 - 269 MB	NPU
FastSam-X	TFLITE	float	Qualcomm® SA8295P	77.396 ms	4 - 343 MB	NPU
FastSam-X	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	25.087 ms	4 - 271 MB	NPU
FastSam-X	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	17.21 ms	4 - 276 MB	NPU

License

• The license for the original implementation of FastSam-X can be found here.

References

• Fast 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.