MobileNet-v2: Optimized for Mobile Deployment

Imagenet classifier and general purpose backbone

MobileNetV2 is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.

This model is an implementation of MobileNet-v2 found here.

This repository provides scripts to run MobileNet-v2 on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

  • Model Type: Model_use_case.image_classification
  • Model Stats:
    • Model checkpoint: Imagenet
    • Input resolution: 224x224
    • Number of parameters: 3.49M
    • Model size (float): 13.3 MB
    • Model size (w8a16): 4.39 MB
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
MobileNet-v2 float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 2.752 ms 0 - 20 MB NPU MobileNet-v2.tflite
MobileNet-v2 float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 2.599 ms 1 - 18 MB NPU MobileNet-v2.dlc
MobileNet-v2 float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 1.033 ms 0 - 31 MB NPU MobileNet-v2.tflite
MobileNet-v2 float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 1.527 ms 0 - 30 MB NPU MobileNet-v2.dlc
MobileNet-v2 float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 0.861 ms 0 - 68 MB NPU MobileNet-v2.tflite
MobileNet-v2 float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 0.834 ms 1 - 47 MB NPU MobileNet-v2.dlc
MobileNet-v2 float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 1.247 ms 0 - 23 MB NPU MobileNet-v2.tflite
MobileNet-v2 float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 1.158 ms 1 - 19 MB NPU MobileNet-v2.dlc
MobileNet-v2 float SA7255P ADP Qualcomm® SA7255P TFLITE 2.752 ms 0 - 20 MB NPU MobileNet-v2.tflite
MobileNet-v2 float SA7255P ADP Qualcomm® SA7255P QNN_DLC 2.599 ms 1 - 18 MB NPU MobileNet-v2.dlc
MobileNet-v2 float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 0.863 ms 0 - 68 MB NPU MobileNet-v2.tflite
MobileNet-v2 float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 0.832 ms 0 - 46 MB NPU MobileNet-v2.dlc
MobileNet-v2 float SA8295P ADP Qualcomm® SA8295P TFLITE 1.437 ms 0 - 22 MB NPU MobileNet-v2.tflite
MobileNet-v2 float SA8295P ADP Qualcomm® SA8295P QNN_DLC 1.383 ms 1 - 21 MB NPU MobileNet-v2.dlc
MobileNet-v2 float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 0.86 ms 0 - 68 MB NPU MobileNet-v2.tflite
MobileNet-v2 float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 0.834 ms 1 - 47 MB NPU MobileNet-v2.dlc
MobileNet-v2 float SA8775P ADP Qualcomm® SA8775P TFLITE 1.247 ms 0 - 23 MB NPU MobileNet-v2.tflite
MobileNet-v2 float SA8775P ADP Qualcomm® SA8775P QNN_DLC 1.158 ms 1 - 19 MB NPU MobileNet-v2.dlc
MobileNet-v2 float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 0.86 ms 0 - 68 MB NPU MobileNet-v2.tflite
MobileNet-v2 float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 0.838 ms 0 - 46 MB NPU MobileNet-v2.dlc
MobileNet-v2 float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 0.716 ms 0 - 24 MB NPU MobileNet-v2.onnx
MobileNet-v2 float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 0.579 ms 0 - 36 MB NPU MobileNet-v2.tflite
MobileNet-v2 float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 0.553 ms 0 - 31 MB NPU MobileNet-v2.dlc
MobileNet-v2 float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 0.485 ms 0 - 34 MB NPU MobileNet-v2.onnx
MobileNet-v2 float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 0.608 ms 0 - 27 MB NPU MobileNet-v2.tflite
MobileNet-v2 float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 0.53 ms 0 - 20 MB NPU MobileNet-v2.dlc
MobileNet-v2 float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 0.51 ms 1 - 24 MB NPU MobileNet-v2.onnx
MobileNet-v2 float Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 1.327 ms 37 - 37 MB NPU MobileNet-v2.dlc
MobileNet-v2 float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 0.79 ms 7 - 7 MB NPU MobileNet-v2.onnx
MobileNet-v2 w8a16 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 1.739 ms 0 - 17 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 0.984 ms 0 - 31 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 0.816 ms 0 - 27 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 1.028 ms 0 - 18 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN_DLC 2.698 ms 0 - 21 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 SA7255P ADP Qualcomm® SA7255P QNN_DLC 1.739 ms 0 - 17 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 0.82 ms 0 - 28 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 SA8295P ADP Qualcomm® SA8295P QNN_DLC 1.226 ms 0 - 23 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 0.822 ms 0 - 29 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 SA8775P ADP Qualcomm® SA8775P QNN_DLC 1.028 ms 0 - 18 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 0.82 ms 0 - 29 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 57.233 ms 10 - 71 MB NPU MobileNet-v2.onnx
MobileNet-v2 w8a16 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 0.567 ms 0 - 31 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 44.723 ms 1 - 396 MB NPU MobileNet-v2.onnx
MobileNet-v2 w8a16 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 0.471 ms 0 - 22 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 38.445 ms 3 - 479 MB NPU MobileNet-v2.onnx
MobileNet-v2 w8a16 Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 1.024 ms 18 - 18 MB NPU MobileNet-v2.dlc
MobileNet-v2 w8a16 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 104.184 ms 26 - 26 MB NPU MobileNet-v2.onnx

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.mobilenet_v2.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.mobilenet_v2.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.mobilenet_v2.export
Profiling Results
------------------------------------------------------------
MobileNet-v2
Device                          : cs_8275 (ANDROID 14)                
Runtime                         : TFLITE                              
Estimated inference time (ms)   : 2.8                                 
Estimated peak memory usage (MB): [0, 20]                             
Total # Ops                     : 71                                  
Compute Unit(s)                 : npu (71 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.mobilenet_v2 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.mobilenet_v2.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.mobilenet_v2.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 MobileNet-v2's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

  • The license for the original implementation of MobileNet-v2 can be found here.
  • The license for the compiled assets for on-device deployment can be found here

References

Community

Downloads last month
165
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for qualcomm/MobileNet-v2

Finetunes
1 model