BiseNet: Optimized for Qualcomm Devices

BiSeNet (Bilateral Segmentation Network) is a novel architecture designed for real-time semantic segmentation. It addresses the challenge of balancing spatial resolution and receptive field by employing a Spatial Path to preserve high-resolution features and a context path to capture sufficient receptive field.

This is based on the implementation of BiseNet 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 2.42, ONNX Runtime 1.24.1	Download
ONNX	w8a8	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
QNN_DLC	w8a8	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.17.0	Download
TFLITE	w8a8	Universal	QAIRT 2.43, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit BiseNet 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 BiseNet on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.semantic_segmentation

Model Stats:

Model checkpoint: best_dice_loss_miou_0.655.pth
Inference latency: RealTime
Input resolution: 720x960
Number of parameters: 12.0M
Model size (float): 45.7 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
BiseNet	ONNX	float	Snapdragon® X Elite	30.848 ms	66 - 66 MB	NPU
BiseNet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	25.504 ms	73 - 318 MB	NPU
BiseNet	ONNX	float	Qualcomm® QCS8550 (Proxy)	32.063 ms	71 - 79 MB	NPU
BiseNet	ONNX	float	Qualcomm® QCS9075	49.312 ms	8 - 11 MB	NPU
BiseNet	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	18.153 ms	65 - 251 MB	NPU
BiseNet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	15.113 ms	56 - 255 MB	NPU
BiseNet	ONNX	float	Snapdragon® X2 Elite	15.48 ms	64 - 64 MB	NPU
BiseNet	ONNX	w8a8	Snapdragon® X Elite	8.661 ms	19 - 19 MB	NPU
BiseNet	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	5.979 ms	18 - 257 MB	NPU
BiseNet	ONNX	w8a8	Qualcomm® QCS6490	239.497 ms	224 - 236 MB	CPU
BiseNet	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	8.315 ms	16 - 22 MB	NPU
BiseNet	ONNX	w8a8	Qualcomm® QCS9075	10.171 ms	18 - 21 MB	NPU
BiseNet	ONNX	w8a8	Qualcomm® QCM6690	229.436 ms	230 - 238 MB	CPU
BiseNet	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	4.752 ms	18 - 218 MB	NPU
BiseNet	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	216.926 ms	250 - 258 MB	CPU
BiseNet	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	3.811 ms	18 - 219 MB	NPU
BiseNet	ONNX	w8a8	Snapdragon® X2 Elite	3.761 ms	17 - 17 MB	NPU
BiseNet	QNN_DLC	float	Snapdragon® X Elite	28.586 ms	8 - 8 MB	NPU
BiseNet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	20.622 ms	8 - 285 MB	NPU
BiseNet	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	107.61 ms	2 - 190 MB	NPU
BiseNet	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	28.399 ms	8 - 10 MB	NPU
BiseNet	QNN_DLC	float	Qualcomm® SA8775P	38.899 ms	1 - 189 MB	NPU
BiseNet	QNN_DLC	float	Qualcomm® QCS9075	55.932 ms	8 - 49 MB	NPU
BiseNet	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	59.788 ms	8 - 277 MB	NPU
BiseNet	QNN_DLC	float	Qualcomm® SA7255P	107.61 ms	2 - 190 MB	NPU
BiseNet	QNN_DLC	float	Qualcomm® SA8295P	44.235 ms	0 - 212 MB	NPU
BiseNet	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	15.769 ms	0 - 222 MB	NPU
BiseNet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	12.716 ms	8 - 286 MB	NPU
BiseNet	QNN_DLC	float	Snapdragon® X2 Elite	14.51 ms	8 - 8 MB	NPU
BiseNet	QNN_DLC	w8a8	Snapdragon® X Elite	10.129 ms	2 - 2 MB	NPU
BiseNet	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	6.686 ms	2 - 231 MB	NPU
BiseNet	QNN_DLC	w8a8	Qualcomm® QCS6490	40.354 ms	1 - 13 MB	NPU
BiseNet	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	20.012 ms	2 - 182 MB	NPU
BiseNet	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	9.516 ms	2 - 4 MB	NPU
BiseNet	QNN_DLC	w8a8	Qualcomm® SA8775P	10.243 ms	2 - 183 MB	NPU
BiseNet	QNN_DLC	w8a8	Qualcomm® QCS9075	13.131 ms	1 - 12 MB	NPU
BiseNet	QNN_DLC	w8a8	Qualcomm® QCM6690	89.126 ms	2 - 206 MB	NPU
BiseNet	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	16.144 ms	2 - 231 MB	NPU
BiseNet	QNN_DLC	w8a8	Qualcomm® SA7255P	20.012 ms	2 - 182 MB	NPU
BiseNet	QNN_DLC	w8a8	Qualcomm® SA8295P	12.698 ms	2 - 185 MB	NPU
BiseNet	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	5.175 ms	2 - 191 MB	NPU
BiseNet	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	13.349 ms	2 - 201 MB	NPU
BiseNet	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	4.297 ms	2 - 193 MB	NPU
BiseNet	QNN_DLC	w8a8	Snapdragon® X2 Elite	4.918 ms	2 - 2 MB	NPU
BiseNet	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	20.464 ms	31 - 288 MB	NPU
BiseNet	TFLITE	float	Qualcomm® QCS8275 (Proxy)	105.803 ms	32 - 247 MB	NPU
BiseNet	TFLITE	float	Qualcomm® QCS8550 (Proxy)	28.692 ms	32 - 34 MB	NPU
BiseNet	TFLITE	float	Qualcomm® SA8775P	37.637 ms	32 - 247 MB	NPU
BiseNet	TFLITE	float	Qualcomm® QCS9075	55.884 ms	0 - 66 MB	NPU
BiseNet	TFLITE	float	Qualcomm® QCS8450 (Proxy)	61.614 ms	32 - 306 MB	NPU
BiseNet	TFLITE	float	Qualcomm® SA7255P	105.803 ms	32 - 247 MB	NPU
BiseNet	TFLITE	float	Qualcomm® SA8295P	44.076 ms	32 - 247 MB	NPU
BiseNet	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	15.878 ms	30 - 253 MB	NPU
BiseNet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	12.63 ms	30 - 309 MB	NPU
BiseNet	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	8.825 ms	6 - 237 MB	NPU
BiseNet	TFLITE	w8a8	Qualcomm® QCS6490	47.441 ms	7 - 31 MB	NPU
BiseNet	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	20.873 ms	8 - 190 MB	NPU
BiseNet	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	11.978 ms	8 - 205 MB	NPU
BiseNet	TFLITE	w8a8	Qualcomm® SA8775P	12.861 ms	8 - 191 MB	NPU
BiseNet	TFLITE	w8a8	Qualcomm® QCS9075	13.153 ms	4 - 28 MB	NPU
BiseNet	TFLITE	w8a8	Qualcomm® QCM6690	92.741 ms	7 - 211 MB	NPU
BiseNet	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	16.253 ms	8 - 238 MB	NPU
BiseNet	TFLITE	w8a8	Qualcomm® SA7255P	20.873 ms	8 - 190 MB	NPU
BiseNet	TFLITE	w8a8	Qualcomm® SA8295P	15.329 ms	8 - 193 MB	NPU
BiseNet	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	6.742 ms	6 - 196 MB	NPU
BiseNet	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	15.908 ms	0 - 199 MB	NPU
BiseNet	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	5.519 ms	6 - 200 MB	NPU

License

The license for the original implementation of BiseNet can be found here.

References

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.

Downloads last month: -; Downloads are not tracked for this model. How to track

Paper for qualcomm/BiseNet

BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation

Paper • 1808.00897 • Published Aug 2, 2018