Segment-Anything-Model-3: Optimized for Qualcomm Devices

SAM3 (Segment Anything with Concepts) extends SAM2 with open-vocabulary segmentation, producing bounding boxes and masks for objects matching a natural-language prompt.

This is based on the implementation of Segment-Anything-Model-3 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

Due to licensing restrictions, we cannot distribute pre-exported model assets for this model. 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

See our repository for Segment-Anything-Model-3 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.semantic_segmentation

Model Stats:

• Model checkpoint: sam3
• Input resolution: 1008x1008
• Number of parameters (SAM3Backbone): 33.5M
• Model size (SAM3Backbone) (float): 128 MB
• Number of parameters (SAM3Transformer): 6.22M
• Model size (SAM3Transformer) (float): 23.7 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
head	PRECOMPILED_QNN_ONNX	float	Snapdragon® X2 Elite	209.595 ms	136 - 136 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Snapdragon® X Elite	489.038 ms	667 - 667 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Gen 3 Mobile	389.841 ms	155 - 168 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Qualcomm® Dragonwing™ IQ-9075	643.219 ms	106 - 215 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	489.038 ms	667 - 667 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Qualcomm® Dragonwing™ Q-8750	282.536 ms	131 - 143 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Mobile	282.536 ms	131 - 143 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	229.69 ms	131 - 144 MB	NPU
head	QNN_CONTEXT_BINARY	float	Snapdragon® X2 Elite	211.571 ms	107 - 107 MB	NPU
head	QNN_CONTEXT_BINARY	float	Snapdragon® X Elite	484.374 ms	106 - 106 MB	NPU
head	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Gen 3 Mobile	384.956 ms	105 - 117 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® Dragonwing™ QCS8275	1006.314 ms	106 - 116 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® SA8775P	517.235 ms	106 - 116 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® SA8650P	517.235 ms	106 - 116 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® SA8255P	517.235 ms	106 - 116 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® Dragonwing™ IQ-9075	502.762 ms	106 - 278 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® Dragonwing™ IQ-X7181	484.374 ms	106 - 106 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® Dragonwing™ Q-8750	270.105 ms	81 - 94 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® SA7255P	1006.314 ms	106 - 116 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® SA8295P	597.053 ms	106 - 111 MB	NPU
head	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Mobile	270.105 ms	81 - 94 MB	NPU
head	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Gen 5 Mobile	242.095 ms	35 - 44 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Snapdragon® X2 Elite	1095.36 ms	265 - 265 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Snapdragon® X Elite	2202.976 ms	943 - 943 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Gen 3 Mobile	1709.651 ms	20 - 31 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Qualcomm® Dragonwing™ IQ-9075	2433.358 ms	12 - 26 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	2202.976 ms	943 - 943 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Qualcomm® Dragonwing™ Q-8750	1427.598 ms	8 - 20 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Mobile	1427.598 ms	8 - 20 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1234.583 ms	28 - 40 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Snapdragon® X2 Elite	1076.267 ms	12 - 12 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Snapdragon® X Elite	2180.381 ms	12 - 12 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Gen 3 Mobile	1687.452 ms	12 - 24 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® Dragonwing™ QCS8275	6386.65 ms	12 - 21 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® SA8775P	2606.394 ms	1 - 10 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® SA8650P	2606.394 ms	1 - 10 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® SA8255P	2606.394 ms	1 - 10 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® Dragonwing™ IQ-9075	2430.088 ms	12 - 131 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® Dragonwing™ IQ-X7181	2180.381 ms	12 - 12 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® Dragonwing™ Q-8750	1412.697 ms	0 - 13 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® SA7255P	6386.65 ms	12 - 21 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® SA8295P	2814.768 ms	1 - 11 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Mobile	1412.697 ms	0 - 13 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Gen 5 Mobile	1211.98 ms	5 - 15 MB	NPU

License

• The license for the original implementation of Segment-Anything-Model-3 can be found here.

References

• SAM 3: Segment Anything with Concepts
• 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.