FFNet-54S-Quantized: Optimized for Mobile Deployment

Semantic segmentation for automotive street scenes

FFNet-54S-Quantized 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 model is an implementation of FFNet-54S-Quantized found here.

This repository provides scripts to run FFNet-54S-Quantized on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

Model Type: Semantic segmentation
Model Stats:
- Model checkpoint: ffnet54S_dBBB_cityscapes_state_dict_quarts
- Input resolution: 2048x1024
- Number of parameters: 18.0M
- Model size: 17.5 MB
- Number of output classes: 19

Model	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Precision	Primary Compute Unit	Target Model
FFNet-54S-Quantized	Samsung Galaxy S23	Snapdragon® 8 Gen 2	TFLITE	4.834 ms	1 - 3 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	Samsung Galaxy S23	Snapdragon® 8 Gen 2	ONNX	10.985 ms	0 - 16 MB	INT8	NPU	FFNet-54S-Quantized.onnx
FFNet-54S-Quantized	Samsung Galaxy S24	Snapdragon® 8 Gen 3	TFLITE	3.358 ms	1 - 73 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	Samsung Galaxy S24	Snapdragon® 8 Gen 3	ONNX	7.693 ms	7 - 129 MB	INT8	NPU	FFNet-54S-Quantized.onnx
FFNet-54S-Quantized	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	TFLITE	3.533 ms	0 - 34 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	ONNX	7.334 ms	0 - 58 MB	INT8	NPU	FFNet-54S-Quantized.onnx
FFNet-54S-Quantized	RB3 Gen 2 (Proxy)	QCS6490 Proxy	TFLITE	31.22 ms	1 - 44 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	RB5 (Proxy)	QCS8250 Proxy	TFLITE	205.499 ms	1 - 3 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	QCS8550 (Proxy)	QCS8550 Proxy	TFLITE	4.693 ms	1 - 3 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	SA8255 (Proxy)	SA8255P Proxy	TFLITE	4.806 ms	1 - 3 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	SA8775 (Proxy)	SA8775P Proxy	TFLITE	4.858 ms	1 - 3 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	SA8650 (Proxy)	SA8650P Proxy	TFLITE	4.712 ms	1 - 3 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	SA8295P ADP	SA8295P	TFLITE	9.219 ms	1 - 34 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	QCS8450 (Proxy)	QCS8450 Proxy	TFLITE	5.94 ms	1 - 76 MB	INT8	NPU	FFNet-54S-Quantized.tflite
FFNet-54S-Quantized	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	11.131 ms	13 - 13 MB	INT8	NPU	FFNet-54S-Quantized.onnx

Installation

This model can be installed as a Python package via pip.

pip install "qai-hub-models[ffnet_54s_quantized]"

Configure Qualcomm® AI Hub to run this model on a cloud-hosted device

Sign-in to Qualcomm® AI Hub with your Qualcomm® ID. Once signed in navigate to Account -> Settings -> API Token.

With this API token, you can configure your client to run models on the cloud hosted devices.

qai-hub configure --api_token API_TOKEN

Navigate to docs for more information.

Demo off target

The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input.

python -m qai_hub_models.models.ffnet_54s_quantized.demo

The above demo runs a reference implementation of pre-processing, model inference, and post processing.

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.ffnet_54s_quantized.demo

Run model on a cloud-hosted device

In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following:

Performance check on-device on a cloud-hosted device
Downloads compiled assets that can be deployed on-device for Android.
Accuracy check between PyTorch and on-device outputs.

python -m qai_hub_models.models.ffnet_54s_quantized.export

Profiling Results
------------------------------------------------------------
FFNet-54S-Quantized
Device                          : Samsung Galaxy S23 (13)
Runtime                         : TFLITE                 
Estimated inference time (ms)   : 4.8                    
Estimated peak memory usage (MB): [1, 3]                 
Total # Ops                     : 120                    
Compute Unit(s)                 : NPU (120 ops)

Run demo on a cloud-hosted device

You can also run the demo on-device.

python -m qai_hub_models.models.ffnet_54s_quantized.demo --on-device

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.ffnet_54s_quantized.demo -- --on-device

Deploying compiled model to Android

The models can be deployed using multiple runtimes:

TensorFlow Lite (.tflite export): This tutorial provides a guide to deploy the .tflite model in an Android application.
QNN (.so export ): This sample app provides instructions on how to use the .so shared library in an Android application.

View on Qualcomm® AI Hub

Get more details on FFNet-54S-Quantized's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of FFNet-54S-Quantized can be found here.
The license for the compiled assets for on-device deployment 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.