ResNet101: Optimized for Mobile Deployment

Imagenet classifier and general purpose backbone

ResNet101 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 ResNet101 found here.

This repository provides scripts to run ResNet101 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: 44.5M
- Model size (float): 170 MB
- Model size (w8a8): 43.9 MB

Model	Precision	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit	Target Model
ResNet101	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	18.389 ms	0 - 87 MB	NPU	ResNet101.tflite
ResNet101	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN	18.292 ms	1 - 10 MB	NPU	Use Export Script
ResNet101	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	4.831 ms	0 - 108 MB	NPU	ResNet101.tflite
ResNet101	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN	5.648 ms	1 - 30 MB	NPU	Use Export Script
ResNet101	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	3.327 ms	0 - 426 MB	NPU	ResNet101.tflite
ResNet101	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN	3.308 ms	1 - 4 MB	NPU	Use Export Script
ResNet101	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	5.544 ms	0 - 87 MB	NPU	ResNet101.tflite
ResNet101	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN	5.431 ms	1 - 13 MB	NPU	Use Export Script
ResNet101	float	SA7255P ADP	Qualcomm® SA7255P	TFLITE	18.389 ms	0 - 87 MB	NPU	ResNet101.tflite
ResNet101	float	SA7255P ADP	Qualcomm® SA7255P	QNN	18.292 ms	1 - 10 MB	NPU	Use Export Script
ResNet101	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	3.344 ms	0 - 304 MB	NPU	ResNet101.tflite
ResNet101	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN	3.284 ms	1 - 4 MB	NPU	Use Export Script
ResNet101	float	SA8295P ADP	Qualcomm® SA8295P	TFLITE	5.646 ms	0 - 78 MB	NPU	ResNet101.tflite
ResNet101	float	SA8295P ADP	Qualcomm® SA8295P	QNN	5.482 ms	1 - 18 MB	NPU	Use Export Script
ResNet101	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	3.361 ms	0 - 335 MB	NPU	ResNet101.tflite
ResNet101	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN	3.302 ms	1 - 3 MB	NPU	Use Export Script
ResNet101	float	SA8775P ADP	Qualcomm® SA8775P	TFLITE	5.544 ms	0 - 87 MB	NPU	ResNet101.tflite
ResNet101	float	SA8775P ADP	Qualcomm® SA8775P	QNN	5.431 ms	1 - 13 MB	NPU	Use Export Script
ResNet101	float	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	TFLITE	3.349 ms	0 - 406 MB	NPU	ResNet101.tflite
ResNet101	float	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	QNN	3.281 ms	1 - 11 MB	NPU	Use Export Script
ResNet101	float	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	ONNX	3.385 ms	0 - 199 MB	NPU	ResNet101.onnx
ResNet101	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	2.411 ms	0 - 115 MB	NPU	ResNet101.tflite
ResNet101	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN	2.408 ms	0 - 41 MB	NPU	Use Export Script
ResNet101	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	2.462 ms	0 - 45 MB	NPU	ResNet101.onnx
ResNet101	float	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	TFLITE	2.295 ms	0 - 92 MB	NPU	ResNet101.tflite
ResNet101	float	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	QNN	2.226 ms	0 - 38 MB	NPU	Use Export Script
ResNet101	float	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	ONNX	2.368 ms	1 - 38 MB	NPU	ResNet101.onnx
ResNet101	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN	3.438 ms	1 - 1 MB	NPU	Use Export Script
ResNet101	float	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	3.407 ms	85 - 85 MB	NPU	ResNet101.onnx
ResNet101	w8a8	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	2.721 ms	0 - 43 MB	NPU	ResNet101.tflite
ResNet101	w8a8	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN	2.943 ms	0 - 10 MB	NPU	Use Export Script
ResNet101	w8a8	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	1.538 ms	0 - 92 MB	NPU	ResNet101.tflite
ResNet101	w8a8	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN	1.956 ms	0 - 77 MB	NPU	Use Export Script
ResNet101	w8a8	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	1.156 ms	0 - 187 MB	NPU	ResNet101.tflite
ResNet101	w8a8	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN	1.258 ms	0 - 4 MB	NPU	Use Export Script
ResNet101	w8a8	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	1.314 ms	0 - 44 MB	NPU	ResNet101.tflite
ResNet101	w8a8	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN	1.441 ms	0 - 13 MB	NPU	Use Export Script
ResNet101	w8a8	RB3 Gen 2 (Proxy)	Qualcomm® QCS6490 (Proxy)	TFLITE	4.552 ms	0 - 76 MB	NPU	ResNet101.tflite
ResNet101	w8a8	RB3 Gen 2 (Proxy)	Qualcomm® QCS6490 (Proxy)	QNN	6.491 ms	0 - 10 MB	NPU	Use Export Script
ResNet101	w8a8	RB5 (Proxy)	Qualcomm® QCS8250 (Proxy)	TFLITE	17.549 ms	0 - 2 MB	NPU	ResNet101.tflite
ResNet101	w8a8	SA7255P ADP	Qualcomm® SA7255P	TFLITE	2.721 ms	0 - 43 MB	NPU	ResNet101.tflite
ResNet101	w8a8	SA7255P ADP	Qualcomm® SA7255P	QNN	2.943 ms	0 - 10 MB	NPU	Use Export Script
ResNet101	w8a8	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	1.155 ms	0 - 190 MB	NPU	ResNet101.tflite
ResNet101	w8a8	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN	1.275 ms	0 - 2 MB	NPU	Use Export Script
ResNet101	w8a8	SA8295P ADP	Qualcomm® SA8295P	TFLITE	1.786 ms	0 - 45 MB	NPU	ResNet101.tflite
ResNet101	w8a8	SA8295P ADP	Qualcomm® SA8295P	QNN	1.868 ms	0 - 18 MB	NPU	Use Export Script
ResNet101	w8a8	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	1.157 ms	0 - 183 MB	NPU	ResNet101.tflite
ResNet101	w8a8	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN	1.271 ms	0 - 2 MB	NPU	Use Export Script
ResNet101	w8a8	SA8775P ADP	Qualcomm® SA8775P	TFLITE	1.314 ms	0 - 44 MB	NPU	ResNet101.tflite
ResNet101	w8a8	SA8775P ADP	Qualcomm® SA8775P	QNN	1.441 ms	0 - 13 MB	NPU	Use Export Script
ResNet101	w8a8	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	TFLITE	1.153 ms	0 - 186 MB	NPU	ResNet101.tflite
ResNet101	w8a8	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	QNN	1.274 ms	0 - 179 MB	NPU	Use Export Script
ResNet101	w8a8	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	ONNX	37.414 ms	10 - 198 MB	NPU	ResNet101.onnx
ResNet101	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	0.885 ms	0 - 91 MB	NPU	ResNet101.tflite
ResNet101	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN	0.975 ms	0 - 74 MB	NPU	Use Export Script
ResNet101	w8a8	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	29.362 ms	13 - 684 MB	NPU	ResNet101.onnx
ResNet101	w8a8	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	TFLITE	0.823 ms	0 - 49 MB	NPU	ResNet101.tflite
ResNet101	w8a8	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	QNN	0.875 ms	1 - 45 MB	NPU	Use Export Script
ResNet101	w8a8	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	ONNX	30.377 ms	10 - 633 MB	NPU	ResNet101.onnx
ResNet101	w8a8	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN	1.246 ms	0 - 0 MB	NPU	Use Export Script
ResNet101	w8a8	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	36.954 ms	55 - 55 MB	NPU	ResNet101.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.resnet101.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.resnet101.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.resnet101.export

Profiling Results
------------------------------------------------------------
ResNet101
Device                          : cs_8275 (ANDROID 14)                 
Runtime                         : TFLITE                               
Estimated inference time (ms)   : 18.4                                 
Estimated peak memory usage (MB): [0, 87]                              
Total # Ops                     : 147                                  
Compute Unit(s)                 : npu (147 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.resnet101 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.resnet101.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.resnet101.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 ResNet101's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of ResNet101 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.