FFNet-54S: Optimized for Qualcomm Devices

FFNet-54S is a "fuss-free network" that segments street scene images with per-pixel classes like road, sidewalk, and pedestrian. Trained on the Cityscapes dataset.

This is based on the implementation of FFNet-54S 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 FFNet-54S 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 FFNet-54S on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.semantic_segmentation

Model Stats:

• Model checkpoint: ffnet54S_dBBB_cityscapes_state_dict_quarts
• Input resolution: 2048x1024
• Number of output classes: 19
• Number of parameters: 18.0M
• Model size (float): 68.8 MB
• Model size (w8a8): 17.5 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
FFNet-54S	ONNX	float	Snapdragon® X2 Elite	14.982 ms	22 - 22 MB	NPU
FFNet-54S	ONNX	float	Snapdragon® X Elite	34.001 ms	24 - 24 MB	NPU
FFNet-54S	ONNX	float	Snapdragon® 8 Gen 3 Mobile	24.367 ms	30 - 315 MB	NPU
FFNet-54S	ONNX	float	Qualcomm® QCS8550 (Proxy)	34.785 ms	24 - 27 MB	NPU
FFNet-54S	ONNX	float	Qualcomm® QCS9075	52.224 ms	24 - 51 MB	NPU
FFNet-54S	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	17.852 ms	7 - 208 MB	NPU
FFNet-54S	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	13.498 ms	15 - 242 MB	NPU
FFNet-54S	ONNX	w8a8	Snapdragon® X2 Elite	7.39 ms	13 - 13 MB	NPU
FFNet-54S	ONNX	w8a8	Snapdragon® X Elite	11.226 ms	12 - 12 MB	NPU
FFNet-54S	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	6.975 ms	7 - 264 MB	NPU
FFNet-54S	ONNX	w8a8	Qualcomm® QCS6490	418.681 ms	182 - 237 MB	CPU
FFNet-54S	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	10.537 ms	0 - 26 MB	NPU
FFNet-54S	ONNX	w8a8	Qualcomm® QCS9075	12.869 ms	6 - 9 MB	NPU
FFNet-54S	ONNX	w8a8	Qualcomm® QCM6690	432.477 ms	148 - 157 MB	CPU
FFNet-54S	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	11.645 ms	1 - 199 MB	NPU
FFNet-54S	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	429.584 ms	201 - 211 MB	CPU
FFNet-54S	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	6.921 ms	7 - 210 MB	NPU
FFNet-54S	QNN_DLC	float	Snapdragon® X2 Elite	15.914 ms	24 - 24 MB	NPU
FFNet-54S	QNN_DLC	float	Snapdragon® X Elite	39.467 ms	24 - 24 MB	NPU
FFNet-54S	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	26.794 ms	14 - 306 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	154.434 ms	0 - 202 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	38.371 ms	24 - 26 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® SA8775P	248.332 ms	24 - 224 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® QCS9075	66.607 ms	24 - 52 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	76.813 ms	24 - 307 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® SA7255P	154.434 ms	0 - 202 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® SA8295P	58.629 ms	24 - 222 MB	NPU
FFNet-54S	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	19.316 ms	16 - 237 MB	NPU
FFNet-54S	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	13.599 ms	6 - 249 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® X2 Elite	6.525 ms	6 - 6 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® X Elite	16.766 ms	6 - 6 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	11.177 ms	6 - 264 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCS6490	71.908 ms	3 - 11 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	35.413 ms	6 - 205 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	15.853 ms	6 - 8 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® SA8775P	16.35 ms	6 - 206 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCS9075	18.921 ms	8 - 16 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCM6690	141.387 ms	6 - 242 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	20.007 ms	6 - 263 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® SA7255P	35.413 ms	6 - 205 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® SA8295P	21.463 ms	6 - 208 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	7.587 ms	6 - 221 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	20.266 ms	6 - 217 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	5.475 ms	6 - 243 MB	NPU
FFNet-54S	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	26.737 ms	2 - 345 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® QCS8275 (Proxy)	154.593 ms	3 - 227 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® QCS8550 (Proxy)	38.02 ms	2 - 5 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® SA8775P	248.278 ms	3 - 226 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® QCS9075	67.005 ms	0 - 64 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® QCS8450 (Proxy)	78.18 ms	0 - 333 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® SA7255P	154.593 ms	3 - 227 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® SA8295P	58.603 ms	2 - 225 MB	NPU
FFNet-54S	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	19.485 ms	2 - 245 MB	NPU
FFNet-54S	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	13.637 ms	2 - 265 MB	NPU
FFNet-54S	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	5.896 ms	1 - 259 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCS6490	55.767 ms	1 - 27 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	22.853 ms	1 - 199 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	8.23 ms	1 - 3 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® SA8775P	8.838 ms	0 - 199 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCS9075	10.055 ms	0 - 26 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCM6690	116.309 ms	1 - 237 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	13.139 ms	1 - 262 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® SA7255P	22.853 ms	1 - 199 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® SA8295P	13.148 ms	1 - 202 MB	NPU
FFNet-54S	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	4.444 ms	1 - 217 MB	NPU
FFNet-54S	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	12.743 ms	1 - 219 MB	NPU
FFNet-54S	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	3.194 ms	1 - 237 MB	NPU

License

• The license for the original implementation of FFNet-54S can be found here.

References

• Simple and Efficient Architectures for Semantic Segmentation
• 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.