ResNet50: Optimized for Mobile Deployment
Imagenet classifier and general purpose backbone
ResNet50 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 ResNet50 found here.
This repository provides scripts to run ResNet50 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: 25.5M
- Model size (float): 97.4 MB
- Model size (w8a8): 25.1 MB
| Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |
|---|---|---|---|---|---|---|---|---|
| ResNet50 | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 10.832 ms | 0 - 66 MB | NPU | ResNet50.tflite |
| ResNet50 | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 10.631 ms | 1 - 34 MB | NPU | ResNet50.dlc |
| ResNet50 | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 3.051 ms | 0 - 76 MB | NPU | ResNet50.tflite |
| ResNet50 | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 3.674 ms | 1 - 37 MB | NPU | ResNet50.dlc |
| ResNet50 | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 2.279 ms | 0 - 355 MB | NPU | ResNet50.tflite |
| ResNet50 | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 2.198 ms | 0 - 15 MB | NPU | ResNet50.dlc |
| ResNet50 | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 2.33 ms | 0 - 141 MB | NPU | ResNet50.onnx.zip |
| ResNet50 | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 3.56 ms | 0 - 66 MB | NPU | ResNet50.tflite |
| ResNet50 | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 3.359 ms | 1 - 34 MB | NPU | ResNet50.dlc |
| ResNet50 | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 10.832 ms | 0 - 66 MB | NPU | ResNet50.tflite |
| ResNet50 | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 10.631 ms | 1 - 34 MB | NPU | ResNet50.dlc |
| ResNet50 | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 2.275 ms | 0 - 351 MB | NPU | ResNet50.tflite |
| ResNet50 | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 2.198 ms | 0 - 21 MB | NPU | ResNet50.dlc |
| ResNet50 | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 3.651 ms | 0 - 59 MB | NPU | ResNet50.tflite |
| ResNet50 | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 3.666 ms | 1 - 31 MB | NPU | ResNet50.dlc |
| ResNet50 | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 2.277 ms | 0 - 358 MB | NPU | ResNet50.tflite |
| ResNet50 | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 2.203 ms | 0 - 23 MB | NPU | ResNet50.dlc |
| ResNet50 | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 3.56 ms | 0 - 66 MB | NPU | ResNet50.tflite |
| ResNet50 | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 3.359 ms | 1 - 34 MB | NPU | ResNet50.dlc |
| ResNet50 | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 1.622 ms | 0 - 82 MB | NPU | ResNet50.tflite |
| ResNet50 | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 1.599 ms | 0 - 42 MB | NPU | ResNet50.dlc |
| ResNet50 | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 1.65 ms | 0 - 39 MB | NPU | ResNet50.onnx.zip |
| ResNet50 | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 1.31 ms | 0 - 69 MB | NPU | ResNet50.tflite |
| ResNet50 | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 1.28 ms | 1 - 40 MB | NPU | ResNet50.dlc |
| ResNet50 | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 1.366 ms | 0 - 37 MB | NPU | ResNet50.onnx.zip |
| ResNet50 | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 1.169 ms | 0 - 68 MB | NPU | ResNet50.tflite |
| ResNet50 | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 1.099 ms | 1 - 58 MB | NPU | ResNet50.dlc |
| ResNet50 | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 1.228 ms | 1 - 59 MB | NPU | ResNet50.onnx.zip |
| ResNet50 | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 2.31 ms | 176 - 176 MB | NPU | ResNet50.dlc |
| ResNet50 | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 2.106 ms | 49 - 49 MB | NPU | ResNet50.onnx.zip |
| ResNet50 | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | TFLITE | 2.674 ms | 0 - 27 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | QNN_DLC | 2.841 ms | 0 - 102 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | ONNX | 32.103 ms | 9 - 30 MB | CPU | ResNet50.onnx.zip |
| ResNet50 | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 1.687 ms | 0 - 34 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 1.939 ms | 0 - 35 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 0.928 ms | 0 - 63 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 1.209 ms | 0 - 63 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 0.776 ms | 0 - 124 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 0.901 ms | 0 - 122 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 1.098 ms | 0 - 117 MB | NPU | ResNet50.onnx.zip |
| ResNet50 | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 0.946 ms | 0 - 34 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 3.876 ms | 0 - 35 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 11.759 ms | 0 - 6 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 21.451 ms | 8 - 37 MB | CPU | ResNet50.onnx.zip |
| ResNet50 | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 1.687 ms | 0 - 34 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 1.939 ms | 0 - 35 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 0.778 ms | 0 - 119 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 0.909 ms | 0 - 125 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 1.249 ms | 0 - 39 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 1.405 ms | 0 - 41 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 0.876 ms | 0 - 121 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 0.903 ms | 0 - 125 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 0.946 ms | 0 - 34 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 3.876 ms | 0 - 35 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.58 ms | 0 - 64 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.677 ms | 0 - 55 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 0.797 ms | 62 - 115 MB | NPU | ResNet50.onnx.zip |
| ResNet50 | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.503 ms | 0 - 46 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 0.53 ms | 0 - 38 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 0.688 ms | 0 - 39 MB | NPU | ResNet50.onnx.zip |
| ResNet50 | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | TFLITE | 1.108 ms | 0 - 49 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | QNN_DLC | 1.256 ms | 0 - 52 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | ONNX | 21.744 ms | 6 - 22 MB | CPU | ResNet50.onnx.zip |
| ResNet50 | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 0.456 ms | 0 - 39 MB | NPU | ResNet50.tflite |
| ResNet50 | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 0.486 ms | 0 - 42 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 0.673 ms | 0 - 41 MB | NPU | ResNet50.onnx.zip |
| ResNet50 | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 0.958 ms | 134 - 134 MB | NPU | ResNet50.dlc |
| ResNet50 | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 0.953 ms | 25 - 25 MB | NPU | ResNet50.onnx.zip |
Installation
Install the package via pip:
pip install qai-hub-models
Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device
Sign-in to Qualcomm® AI Hub Workbench 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.resnet50.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.resnet50.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.resnet50.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.resnet50 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 Workbench. 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.resnet50.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.resnet50.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 ResNet50's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of ResNet50 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.
- Downloads last month
- 404