Hugging Face's logo

qualcomm
/
FFNet-122NS-LowRes

Image Segmentation
PyTorch
LiteRT
ONNX
android
Model card Files Community

FFNet-122NS-LowRes: Optimized for Mobile Deployment

Semantic segmentation for automotive street scenes

FFNet-122NS-LowRes 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-122NS-LowRes found here.

This repository provides scripts to run FFNet-122NS-LowRes 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: ffnet122NS_CCC_cityscapes_state_dict_quarts_pre_down
    • Input resolution: 1024x512
    • Number of output classes: 19
    • Number of parameters: 32.1M
    • Model size (float): 123 MB
    • Model size (w8a8): 31.3 MB
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
FFNet-122NS-LowRes float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 40.514 ms 0 - 60 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 38.241 ms 6 - 35 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 15.283 ms 1 - 118 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 21.879 ms 3 - 43 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 12.305 ms 0 - 44 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 11.523 ms 6 - 17 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 15.553 ms 1 - 61 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 14.613 ms 0 - 29 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float SA7255P ADP Qualcomm® SA7255P TFLITE 40.514 ms 0 - 60 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float SA7255P ADP Qualcomm® SA7255P QNN_DLC 38.241 ms 6 - 35 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 12.301 ms 0 - 44 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 11.587 ms 8 - 17 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float SA8295P ADP Qualcomm® SA8295P TFLITE 17.579 ms 1 - 56 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float SA8295P ADP Qualcomm® SA8295P QNN_DLC 16.754 ms 4 - 32 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 12.389 ms 1 - 14 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 11.561 ms 6 - 19 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float SA8775P ADP Qualcomm® SA8775P TFLITE 15.553 ms 1 - 61 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float SA8775P ADP Qualcomm® SA8775P QNN_DLC 14.613 ms 0 - 29 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 12.345 ms 0 - 91 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 11.535 ms 6 - 18 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 8.67 ms 2 - 183 MB NPU FFNet-122NS-LowRes.onnx
FFNet-122NS-LowRes float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 8.437 ms 1 - 119 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 8.009 ms 6 - 49 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 6.819 ms 8 - 72 MB NPU FFNet-122NS-LowRes.onnx
FFNet-122NS-LowRes float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 7.312 ms 0 - 65 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 7.701 ms 6 - 40 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 6.406 ms 6 - 45 MB NPU FFNet-122NS-LowRes.onnx
FFNet-122NS-LowRes float Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 17.025 ms 97 - 97 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 8.104 ms 57 - 57 MB NPU FFNet-122NS-LowRes.onnx
FFNet-122NS-LowRes w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 6.463 ms 0 - 36 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 9.293 ms 2 - 39 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 3.192 ms 0 - 94 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 6.29 ms 2 - 94 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 2.852 ms 0 - 190 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 4.482 ms 2 - 169 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 3.306 ms 0 - 37 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 4.937 ms 2 - 39 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) TFLITE 9.71 ms 0 - 69 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN_DLC 19.764 ms 0 - 69 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 RB5 (Proxy) Qualcomm® QCS8250 (Proxy) TFLITE 60.6 ms 12 - 24 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 SA7255P ADP Qualcomm® SA7255P TFLITE 6.463 ms 0 - 36 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 SA7255P ADP Qualcomm® SA7255P QNN_DLC 9.293 ms 2 - 39 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 2.837 ms 0 - 189 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 4.487 ms 1 - 172 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 SA8295P ADP Qualcomm® SA8295P TFLITE 4.111 ms 0 - 39 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 SA8295P ADP Qualcomm® SA8295P QNN_DLC 6.06 ms 2 - 43 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 2.845 ms 0 - 190 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 4.483 ms 2 - 140 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 SA8775P ADP Qualcomm® SA8775P TFLITE 3.306 ms 0 - 37 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 SA8775P ADP Qualcomm® SA8775P QNN_DLC 4.937 ms 2 - 39 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 2.861 ms 0 - 188 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 4.49 ms 1 - 172 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 12.558 ms 0 - 70 MB NPU FFNet-122NS-LowRes.onnx
FFNet-122NS-LowRes w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 2.05 ms 0 - 94 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 3.192 ms 2 - 93 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 10.401 ms 8 - 366 MB NPU FFNet-122NS-LowRes.onnx
FFNet-122NS-LowRes w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 1.994 ms 0 - 43 MB NPU FFNet-122NS-LowRes.tflite
FFNet-122NS-LowRes w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 2.563 ms 2 - 41 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 9.448 ms 10 - 570 MB NPU FFNet-122NS-LowRes.onnx
FFNet-122NS-LowRes w8a8 Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 6.338 ms 190 - 190 MB NPU FFNet-122NS-LowRes.dlc
FFNet-122NS-LowRes w8a8 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 15.308 ms 33 - 33 MB NPU FFNet-122NS-LowRes.onnx

Installation

Install the package via pip:

pip install "qai-hub-models[ffnet-122ns-lowres]"

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_122ns_lowres.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_122ns_lowres.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_122ns_lowres.export

Profiling Results
------------------------------------------------------------
FFNet-122NS-LowRes
Device                          : cs_8275 (ANDROID 14)                 
Runtime                         : TFLITE                               
Estimated inference time (ms)   : 40.5                                 
Estimated peak memory usage (MB): [0, 60]                              
Total # Ops                     : 218                                  
Compute Unit(s)                 : npu (218 ops) gpu (0 ops) cpu (0 ops)

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_122ns_lowres import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S24")

# 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. 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_122ns_lowres.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_122ns_lowres.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-122NS-LowRes's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

  • The license for the original implementation of FFNet-122NS-LowRes can be found here.
  • The license for the compiled assets for on-device deployment can be found here

References

Community

Downloads last month
106
Inference Providers NEW
Image Segmentation
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support