YOLOE-Segmentation: Optimized for Qualcomm Devices

Ultralytics YOLOE is a machine learning model that predicts bounding boxes, segmentation masks and classes of objects in an image.

This is based on the implementation of YOLOE-Segmentation 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 YOLOE-Segmentation on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.semantic_segmentation

Model Stats:

• Model checkpoint: YOLOE-v8L-SEG
• Input resolution: 640x640
• Number of parameters: 53.47M
• Model size (float): 103 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
YOLOE-Segmentation	ONNX	float	Snapdragon® X2 Elite	15.676 ms	176 - 176 MB	NPU
YOLOE-Segmentation	ONNX	float	Snapdragon® X Elite	31.024 ms	145 - 145 MB	NPU
YOLOE-Segmentation	ONNX	float	Snapdragon® 8 Gen 3 Mobile	24.169 ms	17 - 289 MB	NPU
YOLOE-Segmentation	ONNX	float	Snapdragon® 8 Gen 1 Mobile	68.04 ms	17 - 335 MB	NPU
YOLOE-Segmentation	ONNX	float	Qualcomm® QCS8550 (Proxy)	30.679 ms	0 - 95 MB	NPU
YOLOE-Segmentation	ONNX	float	Qualcomm® QCS8450	68.04 ms	17 - 335 MB	NPU
YOLOE-Segmentation	ONNX	float	Snapdragon® 8 Elite Mobile	18.194 ms	12 - 232 MB	NPU
YOLOE-Segmentation	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	12.031 ms	13 - 242 MB	NPU
YOLOE-Segmentation	ONNX	float	Qualcomm® QCS9075	48.175 ms	17 - 62 MB	NPU
YOLOE-Segmentation	ONNX	float	Qualcomm® QCS8750	18.194 ms	12 - 232 MB	NPU
YOLOE-Segmentation	ONNX	float	Qualcomm® QCS7181	31.024 ms	145 - 145 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Snapdragon® X2 Elite	15.499 ms	5 - 5 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Snapdragon® X Elite	29.182 ms	5 - 5 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	21.749 ms	5 - 269 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	62.519 ms	5 - 315 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Qualcomm® QCS8275	178.782 ms	1 - 209 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	28.102 ms	5 - 7 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Qualcomm® QCS8450	62.519 ms	5 - 315 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Snapdragon® 8 Elite Mobile	17.151 ms	5 - 219 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Qualcomm® SA7255P	178.782 ms	1 - 209 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Qualcomm® SA8295P	49.982 ms	0 - 248 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	11.406 ms	4 - 216 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Qualcomm® QCS9075	46.076 ms	5 - 15 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Qualcomm® QCS8750	17.151 ms	5 - 219 MB	NPU
YOLOE-Segmentation	QNN_DLC	float	Qualcomm® QCS7181	29.182 ms	5 - 5 MB	NPU
YOLOE-Segmentation	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	21.128 ms	34 - 379 MB	NPU
YOLOE-Segmentation	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	62.578 ms	0 - 391 MB	NPU
YOLOE-Segmentation	TFLITE	float	Qualcomm® QCS8275	177.45 ms	5 - 237 MB	NPU
YOLOE-Segmentation	TFLITE	float	Qualcomm® QCS8550 (Proxy)	27.334 ms	0 - 140 MB	NPU
YOLOE-Segmentation	TFLITE	float	Qualcomm® SA8775P	1634.277 ms	50 - 61 MB	CPU
YOLOE-Segmentation	TFLITE	float	Qualcomm® SA8650P	1634.277 ms	50 - 61 MB	CPU
YOLOE-Segmentation	TFLITE	float	Qualcomm® SA8255P	1634.277 ms	50 - 61 MB	CPU
YOLOE-Segmentation	TFLITE	float	Qualcomm® QCS8450	62.578 ms	0 - 391 MB	NPU
YOLOE-Segmentation	TFLITE	float	Snapdragon® 8 Elite Mobile	16.323 ms	4 - 246 MB	NPU
YOLOE-Segmentation	TFLITE	float	Qualcomm® SA7255P	177.45 ms	5 - 237 MB	NPU
YOLOE-Segmentation	TFLITE	float	Qualcomm® SA8295P	48.524 ms	5 - 277 MB	NPU
YOLOE-Segmentation	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	10.924 ms	3 - 245 MB	NPU
YOLOE-Segmentation	TFLITE	float	Qualcomm® QCS9075	44.982 ms	4 - 106 MB	NPU
YOLOE-Segmentation	TFLITE	float	Qualcomm® QCS8750	16.323 ms	4 - 246 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	6.277 ms	1 - 306 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Snapdragon® 8 Gen 1 Mobile	12.967 ms	1 - 305 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® QCS6490	34.017 ms	1 - 51 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® QCS8275	27.001 ms	1 - 229 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	8.098 ms	1 - 31 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® SA8775P	482.445 ms	13 - 25 MB	CPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® SA8650P	482.445 ms	13 - 25 MB	CPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® SA8255P	482.445 ms	13 - 25 MB	CPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® QCS8450	12.967 ms	1 - 305 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® SA7255P	27.001 ms	1 - 229 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® QCM6690	163.707 ms	1 - 249 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® SA8295P	14.517 ms	1 - 233 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	3.443 ms	1 - 244 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® QCS9075	9.095 ms	0 - 49 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	13.691 ms	0 - 228 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Snapdragon® 8 Elite Mobile	5.001 ms	1 - 232 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® QCS7790	13.691 ms	0 - 228 MB	NPU
YOLOE-Segmentation	TFLITE	w8a8	Qualcomm® QCS8750	5.001 ms	1 - 232 MB	NPU

License

• The license for the original implementation of YOLOE-Segmentation can be found here.

References

• YOLOE: Real-Time Seeing 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.