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	211.492 ms	135 - 135 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Snapdragon® X Elite	484.733 ms	667 - 667 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Gen 3 Mobile	381.125 ms	156 - 167 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	236.184 ms	131 - 143 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Qualcomm® QCS9075	503.413 ms	106 - 215 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Mobile	286.171 ms	132 - 143 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Qualcomm® QCS8750	286.171 ms	132 - 143 MB	NPU
head	PRECOMPILED_QNN_ONNX	float	Qualcomm® QCS7181	484.733 ms	667 - 667 MB	NPU
head	QNN_CONTEXT_BINARY	float	Snapdragon® X2 Elite	210.844 ms	106 - 106 MB	NPU
head	QNN_CONTEXT_BINARY	float	Snapdragon® X Elite	489.544 ms	107 - 107 MB	NPU
head	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Gen 3 Mobile	385.297 ms	106 - 118 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® QCS8275	1006.597 ms	87 - 95 MB	NPU
head	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Gen 5 Mobile	236.135 ms	81 - 90 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® SA7255P	1006.597 ms	87 - 95 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® QCS9075	503.79 ms	106 - 278 MB	NPU
head	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Mobile	279.471 ms	81 - 94 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® SA8295P	596.634 ms	87 - 96 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® QCS8750	279.471 ms	81 - 94 MB	NPU
head	QNN_CONTEXT_BINARY	float	Qualcomm® QCS7181	489.544 ms	107 - 107 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Snapdragon® X2 Elite	1077.436 ms	265 - 265 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Snapdragon® X Elite	2157.567 ms	944 - 944 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Gen 3 Mobile	1721.251 ms	20 - 31 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1208.259 ms	56 - 68 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Qualcomm® QCS9075	2431.819 ms	117 - 131 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Snapdragon® 8 Elite Mobile	1405.929 ms	18 - 29 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Qualcomm® QCS8750	1405.929 ms	18 - 29 MB	NPU
vision_backbone	PRECOMPILED_QNN_ONNX	float	Qualcomm® QCS7181	2157.567 ms	944 - 944 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Snapdragon® X2 Elite	1078.092 ms	12 - 12 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Snapdragon® X Elite	2190.203 ms	12 - 12 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Gen 3 Mobile	1731.302 ms	12 - 27 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® QCS8275	6355.089 ms	8 - 17 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Gen 5 Mobile	1230.139 ms	24 - 33 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® SA7255P	6355.089 ms	8 - 17 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® QCS9075	2430.561 ms	12 - 131 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Snapdragon® 8 Elite Mobile	1397.335 ms	0 - 9 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® SA8295P	2813.386 ms	0 - 9 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® QCS8750	1397.335 ms	0 - 9 MB	NPU
vision_backbone	QNN_CONTEXT_BINARY	float	Qualcomm® QCS7181	2190.203 ms	12 - 12 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.