DeepLabXception: Optimized for Mobile Deployment
Deep Convolutional Neural Network model for semantic segmentation
DeepLabXception is a semantic segmentation model supporting multiple backbones like ResNet-101 and Xception, with flexible dataset compatibility including COCO, VOC, and Cityscapes.
This model is an implementation of DeepLabXception found here.
This repository provides scripts to run DeepLabXception 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: 480x520
- Number of output classes: 21
- Number of parameters: 41.26M
- Model size (float): 158 MB
| Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |
|---|---|---|---|---|---|---|---|---|
| DeepLabXception | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 120.976 ms | 0 - 171 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 113.205 ms | 0 - 78 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 40.449 ms | 0 - 173 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 52.179 ms | 3 - 83 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 24.918 ms | 0 - 24 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 21.9 ms | 3 - 34 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 21.286 ms | 0 - 118 MB | NPU | DeepLabXception.onnx.zip |
| DeepLabXception | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 35.03 ms | 0 - 171 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 140.837 ms | 0 - 78 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 120.976 ms | 0 - 171 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 113.205 ms | 0 - 78 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 24.95 ms | 0 - 23 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 21.877 ms | 3 - 34 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 45.251 ms | 0 - 168 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 39.891 ms | 0 - 76 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 24.899 ms | 0 - 24 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 21.826 ms | 2 - 36 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 35.03 ms | 0 - 171 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 140.837 ms | 0 - 78 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 18.404 ms | 0 - 199 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 16.146 ms | 15 - 123 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 16.059 ms | 3 - 105 MB | NPU | DeepLabXception.onnx.zip |
| DeepLabXception | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 14.553 ms | 0 - 176 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 12.382 ms | 0 - 85 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 12.416 ms | 0 - 79 MB | NPU | DeepLabXception.onnx.zip |
| DeepLabXception | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 11.788 ms | 0 - 175 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 9.613 ms | 3 - 112 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 9.285 ms | 3 - 95 MB | NPU | DeepLabXception.onnx.zip |
| DeepLabXception | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 22.659 ms | 225 - 225 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 21.837 ms | 86 - 86 MB | NPU | DeepLabXception.onnx.zip |
| DeepLabXception | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 20.515 ms | 0 - 111 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 20.774 ms | 1 - 136 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 9.834 ms | 0 - 130 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 15.487 ms | 1 - 144 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 7.588 ms | 0 - 22 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 8.091 ms | 0 - 28 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 8.302 ms | 0 - 121 MB | NPU | DeepLabXception.onnx.zip |
| DeepLabXception | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 8.022 ms | 0 - 111 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 8.323 ms | 1 - 140 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 37.17 ms | 0 - 120 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN_DLC | 54.804 ms | 1 - 137 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | ONNX | 273.616 ms | 90 - 111 MB | CPU | DeepLabXception.onnx.zip |
| DeepLabXception | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 248.672 ms | 19 - 44 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 256.428 ms | 90 - 104 MB | CPU | DeepLabXception.onnx.zip |
| DeepLabXception | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 20.515 ms | 0 - 111 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 20.774 ms | 1 - 136 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 7.598 ms | 0 - 20 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 8.093 ms | 1 - 24 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 12.742 ms | 0 - 115 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 12.845 ms | 1 - 130 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 7.6 ms | 0 - 19 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 8.077 ms | 0 - 26 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 8.022 ms | 0 - 111 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 8.323 ms | 1 - 140 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 5.474 ms | 0 - 140 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 5.729 ms | 0 - 160 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 5.846 ms | 0 - 168 MB | NPU | DeepLabXception.onnx.zip |
| DeepLabXception | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 4.407 ms | 0 - 117 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 4.348 ms | 1 - 127 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 4.635 ms | 1 - 143 MB | NPU | DeepLabXception.onnx.zip |
| DeepLabXception | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 3.351 ms | 0 - 120 MB | NPU | DeepLabXception.tflite |
| DeepLabXception | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 3.281 ms | 1 - 133 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 3.578 ms | 1 - 134 MB | NPU | DeepLabXception.onnx.zip |
| DeepLabXception | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 8.766 ms | 130 - 130 MB | NPU | DeepLabXception.dlc |
| DeepLabXception | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 8.764 ms | 44 - 44 MB | NPU | DeepLabXception.onnx.zip |
Installation
Install the 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.deeplab_xception.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.deeplab_xception.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.deeplab_xception.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.deeplab_xception 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. 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.deeplab_xception.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.deeplab_xception.demo -- --eval-mode on-device
Deploying compiled model to Android
The models can be deployed using multiple runtimes:
TensorFlow Lite (
.tfliteexport): This tutorial provides a guide to deploy the .tflite model in an Android application.QNN (
.soexport ): This sample app provides instructions on how to use the.soshared library in an Android application.
View on Qualcomm® AI Hub
Get more details on DeepLabXception's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of DeepLabXception can be found here.
- The license for the compiled assets for on-device deployment can be found here
References
- Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
- Source Model Implementation
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
- 64