Unet-Segmentation: Optimized for Mobile Deployment
Real-time segmentation optimized for mobile and edge
UNet is a machine learning model that produces a segmentation mask for an image. The most basic use case will label each pixel in the image as being in the foreground or the background. More advanced usage will assign a class label to each pixel. This version of the model was trained on the data from Kaggle's Carvana Image Masking Challenge (see https://www.kaggle.com/c/carvana-image-masking-challenge) and is used for vehicle segmentation.
This model is an implementation of Unet-Segmentation found here.
This repository provides scripts to run Unet-Segmentation 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: unet_carvana_scale1.0_epoch2
- Input resolution: 224x224
- Number of parameters: 31.0M
- Model size: 118 MB
- Number of output classes: 2 (foreground / background)
Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model |
---|---|---|---|---|---|---|---|---|
Unet-Segmentation | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 152.941 ms | 6 - 466 MB | FP16 | NPU | Unet-Segmentation.tflite |
Unet-Segmentation | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 154.481 ms | 9 - 37 MB | FP16 | NPU | Unet-Segmentation.so |
Unet-Segmentation | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 153.838 ms | 16 - 19 MB | FP16 | NPU | Unet-Segmentation.onnx |
Unet-Segmentation | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 113.773 ms | 6 - 92 MB | FP16 | NPU | Unet-Segmentation.tflite |
Unet-Segmentation | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 110.791 ms | 9 - 93 MB | FP16 | NPU | Unet-Segmentation.so |
Unet-Segmentation | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 112.812 ms | 1 - 406 MB | FP16 | NPU | Unet-Segmentation.onnx |
Unet-Segmentation | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 102.108 ms | 4 - 105 MB | FP16 | NPU | Unet-Segmentation.tflite |
Unet-Segmentation | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 89.253 ms | 9 - 110 MB | FP16 | NPU | Use Export Script |
Unet-Segmentation | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 103.52 ms | 14 - 134 MB | FP16 | NPU | Unet-Segmentation.onnx |
Unet-Segmentation | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 157.694 ms | 3 - 469 MB | FP16 | NPU | Unet-Segmentation.tflite |
Unet-Segmentation | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 139.616 ms | 10 - 11 MB | FP16 | NPU | Use Export Script |
Unet-Segmentation | SA7255P ADP | SA7255P | TFLITE | 7406.94 ms | 2 - 100 MB | FP16 | NPU | Unet-Segmentation.tflite |
Unet-Segmentation | SA7255P ADP | SA7255P | QNN | 7399.668 ms | 1 - 7 MB | FP16 | NPU | Use Export Script |
Unet-Segmentation | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 154.76 ms | 6 - 241 MB | FP16 | NPU | Unet-Segmentation.tflite |
Unet-Segmentation | SA8255 (Proxy) | SA8255P Proxy | QNN | 145.08 ms | 10 - 11 MB | FP16 | NPU | Use Export Script |
Unet-Segmentation | SA8295P ADP | SA8295P | TFLITE | 273.606 ms | 6 - 106 MB | FP16 | NPU | Unet-Segmentation.tflite |
Unet-Segmentation | SA8295P ADP | SA8295P | QNN | 266.119 ms | 0 - 6 MB | FP16 | NPU | Use Export Script |
Unet-Segmentation | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 156.964 ms | 3 - 471 MB | FP16 | NPU | Unet-Segmentation.tflite |
Unet-Segmentation | SA8650 (Proxy) | SA8650P Proxy | QNN | 139.333 ms | 10 - 11 MB | FP16 | NPU | Use Export Script |
Unet-Segmentation | SA8775P ADP | SA8775P | TFLITE | 303.207 ms | 6 - 104 MB | FP16 | NPU | Unet-Segmentation.tflite |
Unet-Segmentation | SA8775P ADP | SA8775P | QNN | 297.898 ms | 1 - 6 MB | FP16 | NPU | Use Export Script |
Unet-Segmentation | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 297.003 ms | 6 - 98 MB | FP16 | NPU | Unet-Segmentation.tflite |
Unet-Segmentation | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 322.428 ms | 5 - 92 MB | FP16 | NPU | Use Export Script |
Unet-Segmentation | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 135.772 ms | 9 - 9 MB | FP16 | NPU | Use Export Script |
Unet-Segmentation | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 147.629 ms | 54 - 54 MB | FP16 | NPU | Unet-Segmentation.onnx |
Installation
This model can be installed as a Python package via pip.
pip install qai-hub-models
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.unet_segmentation.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.unet_segmentation.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.unet_segmentation.export
Profiling Results
------------------------------------------------------------
Unet-Segmentation
Device : Samsung Galaxy S23 (13)
Runtime : TFLITE
Estimated inference time (ms) : 152.9
Estimated peak memory usage (MB): [6, 466]
Total # Ops : 32
Compute Unit(s) : NPU (32 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.unet_segmentation import Model
# Load the model
torch_model = Model.from_pretrained()
# Device
device = hub.Device("Samsung Galaxy S23")
# 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.unet_segmentation.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.unet_segmentation.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 Unet-Segmentation's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of Unet-Segmentation 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.