WideResNet50: Optimized for Mobile Deployment

Imagenet classifier and general purpose backbone

WideResNet50 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 WideResNet50 found here.

This repository provides scripts to run WideResNet50 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: 68.8M
    • Model size (float): 263 MB
    • Model size (w8a8): 66.6 MB
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
WideResNet50 float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 23.989 ms 0 - 98 MB NPU WideResNet50.tflite
WideResNet50 float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN 24.312 ms 1 - 10 MB NPU Use Export Script
WideResNet50 float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 6.835 ms 0 - 171 MB NPU WideResNet50.tflite
WideResNet50 float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN 9.749 ms 0 - 40 MB NPU Use Export Script
WideResNet50 float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 4.821 ms 0 - 996 MB NPU WideResNet50.tflite
WideResNet50 float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN 4.816 ms 1 - 4 MB NPU Use Export Script
WideResNet50 float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 7.285 ms 0 - 98 MB NPU WideResNet50.tflite
WideResNet50 float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN 7.09 ms 1 - 10 MB NPU Use Export Script
WideResNet50 float SA7255P ADP Qualcomm® SA7255P TFLITE 23.989 ms 0 - 98 MB NPU WideResNet50.tflite
WideResNet50 float SA7255P ADP Qualcomm® SA7255P QNN 24.312 ms 1 - 10 MB NPU Use Export Script
WideResNet50 float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 4.831 ms 0 - 989 MB NPU WideResNet50.tflite
WideResNet50 float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN 4.796 ms 1 - 3 MB NPU Use Export Script
WideResNet50 float SA8295P ADP Qualcomm® SA8295P TFLITE 7.893 ms 0 - 86 MB NPU WideResNet50.tflite
WideResNet50 float SA8295P ADP Qualcomm® SA8295P QNN 7.671 ms 1 - 17 MB NPU Use Export Script
WideResNet50 float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 4.896 ms 0 - 982 MB NPU WideResNet50.tflite
WideResNet50 float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN 4.824 ms 1 - 3 MB NPU Use Export Script
WideResNet50 float SA8775P ADP Qualcomm® SA8775P TFLITE 7.285 ms 0 - 98 MB NPU WideResNet50.tflite
WideResNet50 float SA8775P ADP Qualcomm® SA8775P QNN 7.09 ms 1 - 10 MB NPU Use Export Script
WideResNet50 float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 4.838 ms 0 - 1026 MB NPU WideResNet50.tflite
WideResNet50 float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN 4.804 ms 1 - 10 MB NPU Use Export Script
WideResNet50 float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 4.828 ms 0 - 295 MB NPU WideResNet50.onnx
WideResNet50 float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 3.574 ms 0 - 173 MB NPU WideResNet50.tflite
WideResNet50 float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN 3.669 ms 1 - 46 MB NPU Use Export Script
WideResNet50 float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 3.611 ms 0 - 49 MB NPU WideResNet50.onnx
WideResNet50 float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 3.377 ms 0 - 103 MB NPU WideResNet50.tflite
WideResNet50 float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN 3.343 ms 1 - 45 MB NPU Use Export Script
WideResNet50 float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 3.433 ms 1 - 46 MB NPU WideResNet50.onnx
WideResNet50 float Snapdragon X Elite CRD Snapdragon® X Elite QNN 4.653 ms 1 - 1 MB NPU Use Export Script
WideResNet50 float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 4.537 ms 132 - 132 MB NPU WideResNet50.onnx
WideResNet50 w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 3.754 ms 0 - 37 MB NPU WideResNet50.tflite
WideResNet50 w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN 4.015 ms 0 - 9 MB NPU Use Export Script
WideResNet50 w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 2.201 ms 0 - 103 MB NPU WideResNet50.tflite
WideResNet50 w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN 2.526 ms 0 - 103 MB NPU Use Export Script
WideResNet50 w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 1.771 ms 0 - 383 MB NPU WideResNet50.tflite
WideResNet50 w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN 1.869 ms 0 - 4 MB NPU Use Export Script
WideResNet50 w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 1.919 ms 0 - 37 MB NPU WideResNet50.tflite
WideResNet50 w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN 1.989 ms 0 - 10 MB NPU Use Export Script
WideResNet50 w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) TFLITE 7.356 ms 0 - 93 MB NPU WideResNet50.tflite
WideResNet50 w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN 9.693 ms 0 - 13 MB NPU Use Export Script
WideResNet50 w8a8 RB5 (Proxy) Qualcomm® QCS8250 (Proxy) TFLITE 23.535 ms 0 - 2 MB NPU WideResNet50.tflite
WideResNet50 w8a8 SA7255P ADP Qualcomm® SA7255P TFLITE 3.754 ms 0 - 37 MB NPU WideResNet50.tflite
WideResNet50 w8a8 SA7255P ADP Qualcomm® SA7255P QNN 4.015 ms 0 - 9 MB NPU Use Export Script
WideResNet50 w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 1.78 ms 0 - 384 MB NPU WideResNet50.tflite
WideResNet50 w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN 1.87 ms 0 - 7 MB NPU Use Export Script
WideResNet50 w8a8 SA8295P ADP Qualcomm® SA8295P TFLITE 2.605 ms 0 - 39 MB NPU WideResNet50.tflite
WideResNet50 w8a8 SA8295P ADP Qualcomm® SA8295P QNN 2.785 ms 0 - 17 MB NPU Use Export Script
WideResNet50 w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 1.777 ms 0 - 387 MB NPU WideResNet50.tflite
WideResNet50 w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN 1.87 ms 0 - 2 MB NPU Use Export Script
WideResNet50 w8a8 SA8775P ADP Qualcomm® SA8775P TFLITE 1.919 ms 0 - 37 MB NPU WideResNet50.tflite
WideResNet50 w8a8 SA8775P ADP Qualcomm® SA8775P QNN 1.989 ms 0 - 10 MB NPU Use Export Script
WideResNet50 w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 1.785 ms 0 - 382 MB NPU WideResNet50.tflite
WideResNet50 w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN 1.865 ms 0 - 367 MB NPU Use Export Script
WideResNet50 w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 18.051 ms 2 - 136 MB NPU WideResNet50.onnx
WideResNet50 w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 1.351 ms 0 - 104 MB NPU WideResNet50.tflite
WideResNet50 w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN 1.407 ms 0 - 100 MB NPU Use Export Script
WideResNet50 w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 13.978 ms 0 - 502 MB NPU WideResNet50.onnx
WideResNet50 w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 1.247 ms 0 - 43 MB NPU WideResNet50.tflite
WideResNet50 w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN 1.294 ms 0 - 45 MB NPU Use Export Script
WideResNet50 w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 14.503 ms 2 - 614 MB NPU WideResNet50.onnx
WideResNet50 w8a8 Snapdragon X Elite CRD Snapdragon® X Elite QNN 1.793 ms 0 - 0 MB NPU Use Export Script
WideResNet50 w8a8 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 17.323 ms 76 - 76 MB NPU WideResNet50.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.wideresnet50.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.wideresnet50.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.wideresnet50.export
Profiling Results
------------------------------------------------------------
WideResNet50
Device                          : cs_8275 (ANDROID 14)                
Runtime                         : TFLITE                              
Estimated inference time (ms)   : 24.0                                
Estimated peak memory usage (MB): [0, 98]                             
Total # Ops                     : 79                                  
Compute Unit(s)                 : npu (79 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.wideresnet50 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.wideresnet50.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.wideresnet50.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 WideResNet50's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

  • The license for the original implementation of WideResNet50 can be found here.
  • The license for the compiled assets for on-device deployment can be found here

References

Community

Downloads last month
52
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support