Hugging Face's logo

qualcomm
/
FFNet-54S

Image Segmentation
PyTorch
LiteRT
android
Model card Files
xet
Community

FFNet-54S: Optimized for Mobile Deployment

Semantic segmentation for automotive street scenes

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

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

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
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
FFNet-54S float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 152.164 ms 3 - 197 MB NPU FFNet-54S.tflite
FFNet-54S float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 152.355 ms 24 - 197 MB NPU FFNet-54S.dlc
FFNet-54S float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 70.096 ms 3 - 307 MB NPU FFNet-54S.tflite
FFNet-54S float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 70.244 ms 24 - 275 MB NPU FFNet-54S.dlc
FFNet-54S float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 36.866 ms 2 - 5 MB NPU FFNet-54S.tflite
FFNet-54S float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 37.106 ms 18 - 20 MB NPU FFNet-54S.dlc
FFNet-54S float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) ONNX 30.586 ms 24 - 39 MB NPU FFNet-54S.onnx.zip
FFNet-54S float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 51.766 ms 0 - 194 MB NPU FFNet-54S.tflite
FFNet-54S float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 51.751 ms 24 - 196 MB NPU FFNet-54S.dlc
FFNet-54S float SA7255P ADP Qualcomm® SA7255P TFLITE 152.164 ms 3 - 197 MB NPU FFNet-54S.tflite
FFNet-54S float SA7255P ADP Qualcomm® SA7255P QNN_DLC 152.355 ms 24 - 197 MB NPU FFNet-54S.dlc
FFNet-54S float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 36.981 ms 2 - 4 MB NPU FFNet-54S.tflite
FFNet-54S float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 37.06 ms 24 - 26 MB NPU FFNet-54S.dlc
FFNet-54S float SA8295P ADP Qualcomm® SA8295P TFLITE 56.655 ms 2 - 195 MB NPU FFNet-54S.tflite
FFNet-54S float SA8295P ADP Qualcomm® SA8295P QNN_DLC 56.634 ms 24 - 194 MB NPU FFNet-54S.dlc
FFNet-54S float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 36.914 ms 2 - 6 MB NPU FFNet-54S.tflite
FFNet-54S float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 36.671 ms 22 - 24 MB NPU FFNet-54S.dlc
FFNet-54S float SA8775P ADP Qualcomm® SA8775P TFLITE 51.766 ms 0 - 194 MB NPU FFNet-54S.tflite
FFNet-54S float SA8775P ADP Qualcomm® SA8775P QNN_DLC 51.751 ms 24 - 196 MB NPU FFNet-54S.dlc
FFNet-54S float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 25.101 ms 2 - 312 MB NPU FFNet-54S.tflite
FFNet-54S float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 25.125 ms 24 - 286 MB NPU FFNet-54S.dlc
FFNet-54S float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 21.066 ms 32 - 258 MB NPU FFNet-54S.onnx.zip
FFNet-54S float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile TFLITE 18.821 ms 1 - 210 MB NPU FFNet-54S.tflite
FFNet-54S float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile QNN_DLC 18.867 ms 8 - 197 MB NPU FFNet-54S.dlc
FFNet-54S float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile ONNX 15.679 ms 7 - 155 MB NPU FFNet-54S.onnx.zip
FFNet-54S float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile TFLITE 13.854 ms 2 - 222 MB NPU FFNet-54S.tflite
FFNet-54S float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile QNN_DLC 13.799 ms 24 - 227 MB NPU FFNet-54S.dlc
FFNet-54S float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile ONNX 11.747 ms 29 - 195 MB NPU FFNet-54S.onnx.zip
FFNet-54S float Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 37.895 ms 24 - 24 MB NPU FFNet-54S.dlc
FFNet-54S float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 30.415 ms 24 - 24 MB NPU FFNet-54S.onnx.zip
FFNet-54S w8a8 Dragonwing Q-6690 MTP Qualcomm® Qcm6690 TFLITE 111.914 ms 1 - 211 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 Dragonwing Q-6690 MTP Qualcomm® Qcm6690 QNN_DLC 123.049 ms 6 - 218 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 Dragonwing Q-6690 MTP Qualcomm® Qcm6690 ONNX 436.528 ms 141 - 156 MB CPU FFNet-54S.onnx.zip
FFNet-54S w8a8 Dragonwing RB3 Gen 2 Vision Kit Qualcomm® QCS6490 TFLITE 45.868 ms 1 - 27 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 Dragonwing RB3 Gen 2 Vision Kit Qualcomm® QCS6490 QNN_DLC 67.779 ms 6 - 14 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 Dragonwing RB3 Gen 2 Vision Kit Qualcomm® QCS6490 ONNX 408.577 ms 179 - 236 MB CPU FFNet-54S.onnx.zip
FFNet-54S w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 22.734 ms 1 - 172 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 35.844 ms 6 - 177 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 13.046 ms 1 - 233 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 19.958 ms 6 - 234 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 8.178 ms 1 - 4 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 15.974 ms 6 - 8 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) ONNX 10.388 ms 0 - 16 MB NPU FFNet-54S.onnx.zip
FFNet-54S w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 8.824 ms 1 - 171 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 74.145 ms 6 - 177 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 RB5 (Proxy) Qualcomm® QCS8250 (Proxy) TFLITE 203.478 ms 8 - 95 MB GPU FFNet-54S.tflite
FFNet-54S w8a8 RB5 (Proxy) Qualcomm® QCS8250 (Proxy) ONNX 365.501 ms 166 - 186 MB CPU FFNet-54S.onnx.zip
FFNet-54S w8a8 SA7255P ADP Qualcomm® SA7255P TFLITE 22.734 ms 1 - 172 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 SA7255P ADP Qualcomm® SA7255P QNN_DLC 35.844 ms 6 - 177 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 8.201 ms 1 - 7 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 15.921 ms 6 - 8 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 SA8295P ADP Qualcomm® SA8295P TFLITE 13.004 ms 1 - 175 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 SA8295P ADP Qualcomm® SA8295P QNN_DLC 21.646 ms 6 - 192 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 8.216 ms 1 - 3 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 15.976 ms 6 - 8 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 SA8775P ADP Qualcomm® SA8775P TFLITE 8.824 ms 1 - 171 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 SA8775P ADP Qualcomm® SA8775P QNN_DLC 74.145 ms 6 - 177 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 5.821 ms 1 - 233 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 11.068 ms 6 - 234 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 7.23 ms 7 - 211 MB NPU FFNet-54S.onnx.zip
FFNet-54S w8a8 Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile TFLITE 4.444 ms 0 - 179 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile QNN_DLC 7.47 ms 6 - 190 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile ONNX 7.197 ms 1 - 145 MB NPU FFNet-54S.onnx.zip
FFNet-54S w8a8 Snapdragon 7 Gen 4 QRD Snapdragon® 7 Gen 4 Mobile TFLITE 12.142 ms 1 - 181 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 Snapdragon 7 Gen 4 QRD Snapdragon® 7 Gen 4 Mobile QNN_DLC 19.943 ms 6 - 197 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 Snapdragon 7 Gen 4 QRD Snapdragon® 7 Gen 4 Mobile ONNX 426.65 ms 193 - 211 MB CPU FFNet-54S.onnx.zip
FFNet-54S w8a8 Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile TFLITE 3.197 ms 1 - 201 MB NPU FFNet-54S.tflite
FFNet-54S w8a8 Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile QNN_DLC 5.474 ms 6 - 207 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile ONNX 4.074 ms 9 - 158 MB NPU FFNet-54S.onnx.zip
FFNet-54S w8a8 Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 16.758 ms 6 - 6 MB NPU FFNet-54S.dlc
FFNet-54S w8a8 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 11.2 ms 13 - 13 MB NPU FFNet-54S.onnx.zip

Installation

Install the package via pip:

# NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported.
pip install "qai-hub-models[ffnet-54s]"

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

Sign-in to Qualcomm® AI Hub Workbench 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.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.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.export

How does this work?

This export script leverages Qualcomm® AI Hub to optimize, validate, and deploy this model on-device. Lets go through each step below in detail:

Step 1: Compile model for on-device deployment

To compile a PyTorch model for on-device deployment, we first trace the model in memory using the jit.trace and then call the submit_compile_job API.

import torch

import qai_hub as hub
from qai_hub_models.models.ffnet_54s import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S25")

# Trace model
input_shape = torch_model.get_input_spec()
sample_inputs = torch_model.sample_inputs()

pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])

# Compile model on a specific device
compile_job = hub.submit_compile_job(
    model=pt_model,
    device=device,
    input_specs=torch_model.get_input_spec(),
)

# Get target model to run on-device
target_model = compile_job.get_target_model()

Step 2: Performance profiling on cloud-hosted device

After compiling models from step 1. Models can be profiled model on-device using the target_model. Note that this scripts runs the model on a device automatically provisioned in the cloud. Once the job is submitted, you can navigate to a provided job URL to view a variety of on-device performance metrics.

profile_job = hub.submit_profile_job(
    model=target_model,
    device=device,
)

Step 3: Verify on-device accuracy

To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device.

input_data = torch_model.sample_inputs()
inference_job = hub.submit_inference_job(
    model=target_model,
    device=device,
    inputs=input_data,
)
    on_device_output = inference_job.download_output_data()

With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output.

Note: This on-device profiling and inference requires access to Qualcomm® AI Hub Workbench. Sign up for access.

Run demo on a cloud-hosted device

You can also run the demo on-device.

python -m qai_hub_models.models.ffnet_54s.demo --eval-mode 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.demo -- --eval-mode 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's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

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

References

Community

Downloads last month
244
Inference Providers NEW
Image Segmentation
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Paper for qualcomm/FFNet-54S