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 2.37, ONNX Runtime 1.23.0	Download
ONNX	w8a8	Universal	QAIRT 2.37, ONNX Runtime 1.23.0	Download
QNN_DLC	float	Universal	QAIRT 2.42	Download
QNN_DLC	w8a8	Universal	QAIRT 2.42	Download
TFLITE	float	Universal	QAIRT 2.42, TFLite 2.17.0	Download
TFLITE	w8a8	Universal	QAIRT 2.42, 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® X Elite	30.339 ms	24 - 24 MB	NPU
FFNet-54S	ONNX	float	Snapdragon® 8 Gen 3 Mobile	20.752 ms	0 - 229 MB	NPU
FFNet-54S	ONNX	float	Qualcomm® QCS8550 (Proxy)	30.892 ms	24 - 43 MB	NPU
FFNet-54S	ONNX	float	Qualcomm® QCS9075	52.843 ms	24 - 51 MB	NPU
FFNet-54S	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	15.729 ms	7 - 157 MB	NPU
FFNet-54S	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	11.752 ms	12 - 180 MB	NPU
FFNet-54S	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	7.267 ms	7 - 212 MB	NPU
FFNet-54S	ONNX	w8a8	Qualcomm® QCS6490	408.641 ms	179 - 236 MB	CPU
FFNet-54S	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	10.524 ms	0 - 16 MB	NPU
FFNet-54S	ONNX	w8a8	Qualcomm® QCS9075	13.397 ms	6 - 9 MB	NPU
FFNet-54S	ONNX	w8a8	Qualcomm® QCM6690	435.263 ms	149 - 158 MB	CPU
FFNet-54S	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	7.198 ms	1 - 149 MB	NPU
FFNet-54S	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	428.029 ms	201 - 211 MB	CPU
FFNet-54S	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	4.078 ms	0 - 151 MB	NPU
FFNet-54S	QNN_DLC	float	Snapdragon® X Elite	39.954 ms	24 - 24 MB	NPU
FFNet-54S	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	27.098 ms	23 - 320 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	155.821 ms	24 - 229 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	38.882 ms	24 - 26 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® SA8775P	53.647 ms	24 - 226 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® QCS9075	66.953 ms	24 - 52 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	77.387 ms	24 - 305 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® SA7255P	155.821 ms	24 - 229 MB	NPU
FFNet-54S	QNN_DLC	float	Qualcomm® SA8295P	59.293 ms	24 - 225 MB	NPU
FFNet-54S	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	19.597 ms	18 - 248 MB	NPU
FFNet-54S	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	13.695 ms	10 - 255 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® X Elite	16.756 ms	6 - 6 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	11.186 ms	6 - 264 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCS6490	72.165 ms	6 - 14 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	35.546 ms	6 - 205 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	15.85 ms	6 - 8 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® SA8775P	16.336 ms	6 - 206 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCS9075	18.463 ms	6 - 14 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCM6690	142.721 ms	6 - 243 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	19.94 ms	6 - 265 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® SA7255P	35.546 ms	6 - 205 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Qualcomm® SA8295P	21.512 ms	6 - 208 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	7.538 ms	6 - 221 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	20.365 ms	6 - 217 MB	NPU
FFNet-54S	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	5.45 ms	6 - 244 MB	NPU
FFNet-54S	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	27.169 ms	1 - 349 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® QCS8275 (Proxy)	155.903 ms	3 - 230 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® QCS8550 (Proxy)	38.37 ms	2 - 5 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® SA8775P	248.039 ms	3 - 231 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® QCS9075	66.642 ms	0 - 64 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® QCS8450 (Proxy)	77.922 ms	3 - 335 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® SA7255P	155.903 ms	3 - 230 MB	NPU
FFNet-54S	TFLITE	float	Qualcomm® SA8295P	59.309 ms	2 - 226 MB	NPU
FFNet-54S	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	19.682 ms	0 - 246 MB	NPU
FFNet-54S	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	13.629 ms	2 - 265 MB	NPU
FFNet-54S	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	5.898 ms	1 - 260 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCS6490	55.702 ms	1 - 27 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	22.888 ms	1 - 198 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	8.24 ms	1 - 4 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® SA8775P	8.852 ms	1 - 200 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCS9075	10.012 ms	0 - 26 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCM6690	117.556 ms	1 - 237 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	13.081 ms	1 - 261 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® SA7255P	22.888 ms	1 - 198 MB	NPU
FFNet-54S	TFLITE	w8a8	Qualcomm® SA8295P	13.203 ms	1 - 203 MB	NPU
FFNet-54S	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	4.448 ms	0 - 214 MB	NPU
FFNet-54S	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	12.677 ms	1 - 217 MB	NPU
FFNet-54S	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	3.198 ms	1 - 236 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.