FCN-ResNet50: Optimized for Mobile Deployment

Fully-convolutional network model for image segmentation

FCN_ResNet50 is a machine learning model that can segment images from the COCO dataset. It uses ResNet50 as a backbone.

This model is an implementation of FCN-ResNet50 found here.

This repository provides scripts to run FCN-ResNet50 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: COCO_WITH_VOC_LABELS_V1
- Input resolution: 224x224
- Number of output classes: 21
- Number of parameters: 33.0M
- Model size (float): 126 MB
- Model size (w8a8): 32.2 MB

Model	Precision	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit	Target Model
FCN-ResNet50	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	277.878 ms	0 - 140 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN_DLC	271.268 ms	82 - 213 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	74.62 ms	0 - 121 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN_DLC	102.285 ms	3 - 101 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	47.589 ms	2 - 23 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_DLC	43.558 ms	3 - 48 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	ONNX	44.907 ms	4 - 45 MB	NPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	77.533 ms	0 - 139 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_DLC	72.423 ms	0 - 129 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	SA7255P ADP	Qualcomm® SA7255P	TFLITE	277.878 ms	0 - 140 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	SA7255P ADP	Qualcomm® SA7255P	QNN_DLC	271.268 ms	82 - 213 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	47.518 ms	0 - 25 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_DLC	43.254 ms	3 - 48 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	SA8295P ADP	Qualcomm® SA8295P	TFLITE	84.782 ms	0 - 97 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	SA8295P ADP	Qualcomm® SA8295P	QNN_DLC	93.588 ms	1 - 104 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	47.467 ms	0 - 21 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_DLC	43.612 ms	2 - 47 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	SA8775P ADP	Qualcomm® SA8775P	TFLITE	77.533 ms	0 - 139 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	SA8775P ADP	Qualcomm® SA8775P	QNN_DLC	72.423 ms	0 - 129 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	35.122 ms	0 - 160 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_DLC	32.315 ms	3 - 141 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	32.826 ms	3 - 137 MB	NPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	TFLITE	29.166 ms	0 - 143 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_DLC	26.556 ms	2 - 153 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	ONNX	26.945 ms	0 - 126 MB	NPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	TFLITE	24.357 ms	0 - 146 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_DLC	23.926 ms	3 - 160 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	ONNX	21.613 ms	3 - 140 MB	NPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_DLC	43.628 ms	39 - 39 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	float	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	43.665 ms	63 - 63 MB	NPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	w8a8	Dragonwing RB3 Gen 2 Vision Kit	Qualcomm® QCS6490	TFLITE	80.929 ms	0 - 39 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	Dragonwing RB3 Gen 2 Vision Kit	Qualcomm® QCS6490	QNN_DLC	93.168 ms	0 - 194 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	Dragonwing RB3 Gen 2 Vision Kit	Qualcomm® QCS6490	ONNX	900.399 ms	66 - 114 MB	CPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	w8a8	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	39.484 ms	0 - 54 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN_DLC	38.398 ms	1 - 72 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	17.509 ms	0 - 87 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN_DLC	23.703 ms	1 - 99 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	14.62 ms	0 - 13 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_DLC	14.149 ms	1 - 20 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	ONNX	13.84 ms	0 - 96 MB	NPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	w8a8	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	15.104 ms	0 - 54 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_DLC	14.633 ms	1 - 71 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	RB5 (Proxy)	Qualcomm® QCS8250 (Proxy)	TFLITE	1315.116 ms	69 - 149 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	RB5 (Proxy)	Qualcomm® QCS8250 (Proxy)	ONNX	787.261 ms	64 - 124 MB	CPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	w8a8	SA7255P ADP	Qualcomm® SA7255P	TFLITE	39.484 ms	0 - 54 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	SA7255P ADP	Qualcomm® SA7255P	QNN_DLC	38.398 ms	1 - 72 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	14.66 ms	0 - 12 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_DLC	14.182 ms	0 - 20 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	SA8295P ADP	Qualcomm® SA8295P	TFLITE	21.846 ms	0 - 58 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	SA8295P ADP	Qualcomm® SA8295P	QNN_DLC	21.032 ms	1 - 76 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	14.676 ms	0 - 14 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_DLC	14.203 ms	1 - 20 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	SA8775P ADP	Qualcomm® SA8775P	TFLITE	15.104 ms	0 - 54 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	SA8775P ADP	Qualcomm® SA8775P	QNN_DLC	14.633 ms	1 - 71 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	10.895 ms	0 - 90 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_DLC	10.579 ms	1 - 99 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	10.199 ms	1 - 113 MB	NPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	w8a8	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	TFLITE	9.087 ms	0 - 59 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_DLC	8.675 ms	1 - 75 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	ONNX	8.696 ms	1 - 72 MB	NPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	w8a8	Snapdragon 7 Gen 4 QRD	Snapdragon® 7 Gen 4 Mobile	TFLITE	28.019 ms	0 - 84 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	Snapdragon 7 Gen 4 QRD	Snapdragon® 7 Gen 4 Mobile	QNN_DLC	27.903 ms	1 - 129 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	Snapdragon 7 Gen 4 QRD	Snapdragon® 7 Gen 4 Mobile	ONNX	698.872 ms	52 - 68 MB	CPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	w8a8	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	TFLITE	7.968 ms	0 - 58 MB	NPU	FCN-ResNet50.tflite
FCN-ResNet50	w8a8	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_DLC	7.308 ms	1 - 98 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	ONNX	7.32 ms	1 - 97 MB	NPU	FCN-ResNet50.onnx.zip
FCN-ResNet50	w8a8	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_DLC	14.488 ms	119 - 119 MB	NPU	FCN-ResNet50.dlc
FCN-ResNet50	w8a8	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	13.911 ms	32 - 32 MB	NPU	FCN-ResNet50.onnx.zip

Installation

Install the package via pip:

pip install qai-hub-models

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.fcn_resnet50.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.fcn_resnet50.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.fcn_resnet50.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.fcn_resnet50 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.fcn_resnet50.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.fcn_resnet50.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 FCN-ResNet50's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of FCN-ResNet50 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.

Downloads last month: 428