---
library_name: pytorch
license: other
tags:
- backbone
- bu_auto
- android
pipeline_tag: image-classification

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet18/web-assets/model_demo.png)

# ResNet18: Optimized for Qualcomm Devices

ResNet18 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 is based on the implementation of ResNet18 found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/resnet18) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.

## Getting Started
There are two ways to deploy this model on your device:

### Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

| Runtime | Precision | Chipset | SDK Versions | Download |
|---|---|---|---|---|
| ONNX | float | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet18/releases/v0.46.0/resnet18-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet18/releases/v0.46.0/resnet18-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet18/releases/v0.46.0/resnet18-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet18/releases/v0.46.0/resnet18-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet18/releases/v0.46.0/resnet18-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet18/releases/v0.46.0/resnet18-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[ResNet18 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/resnet18)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/resnet18) Python library to compile and export the model with your own:
- Custom weights (e.g., fine-tuned checkpoints)
- Custom input shapes
- Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for [ResNet18 on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/resnet18) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Model checkpoint: Imagenet
- Input resolution: 224x224
- Number of parameters: 11.7M
- Model size (float): 44.6 MB
- Model size (w8a8): 11.3 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| ResNet18 | ONNX | float | Snapdragon® X Elite | 1.151 ms | 23 - 23 MB | NPU
| ResNet18 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.939 ms | 0 - 103 MB | NPU
| ResNet18 | ONNX | float | Qualcomm® QCS8550 (Proxy) | 1.237 ms | 0 - 25 MB | NPU
| ResNet18 | ONNX | float | Qualcomm® QCS9075 | 2.0 ms | 1 - 3 MB | NPU
| ResNet18 | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.75 ms | 0 - 90 MB | NPU
| ResNet18 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.657 ms | 0 - 91 MB | NPU
| ResNet18 | ONNX | w8a8 | Snapdragon® X Elite | 0.613 ms | 11 - 11 MB | NPU
| ResNet18 | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.582 ms | 0 - 116 MB | NPU
| ResNet18 | ONNX | w8a8 | Qualcomm® QCS6490 | 13.761 ms | 6 - 24 MB | CPU
| ResNet18 | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.747 ms | 0 - 152 MB | NPU
| ResNet18 | ONNX | w8a8 | Qualcomm® QCS9075 | 0.858 ms | 0 - 3 MB | NPU
| ResNet18 | ONNX | w8a8 | Qualcomm® QCM6690 | 11.209 ms | 6 - 14 MB | CPU
| ResNet18 | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.476 ms | 0 - 92 MB | NPU
| ResNet18 | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 8.586 ms | 8 - 15 MB | CPU
| ResNet18 | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.435 ms | 0 - 94 MB | NPU
| ResNet18 | QNN_DLC | float | Snapdragon® X Elite | 1.435 ms | 1 - 1 MB | NPU
| ResNet18 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.929 ms | 0 - 34 MB | NPU
| ResNet18 | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 6.004 ms | 1 - 24 MB | NPU
| ResNet18 | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 1.319 ms | 1 - 145 MB | NPU
| ResNet18 | QNN_DLC | float | Qualcomm® SA8775P | 1.981 ms | 1 - 27 MB | NPU
| ResNet18 | QNN_DLC | float | Qualcomm® QCS9075 | 2.059 ms | 1 - 3 MB | NPU
| ResNet18 | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 2.432 ms | 0 - 37 MB | NPU
| ResNet18 | QNN_DLC | float | Qualcomm® SA7255P | 6.004 ms | 1 - 24 MB | NPU
| ResNet18 | QNN_DLC | float | Qualcomm® SA8295P | 2.342 ms | 1 - 20 MB | NPU
| ResNet18 | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.72 ms | 0 - 23 MB | NPU
| ResNet18 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.594 ms | 1 - 27 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.622 ms | 0 - 0 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.405 ms | 0 - 48 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 1.811 ms | 0 - 2 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 1.309 ms | 0 - 24 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.548 ms | 0 - 2 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.708 ms | 0 - 25 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 0.61 ms | 0 - 2 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 3.239 ms | 0 - 31 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.743 ms | 0 - 48 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Qualcomm® SA7255P | 1.309 ms | 0 - 24 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Qualcomm® SA8295P | 0.957 ms | 0 - 21 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.299 ms | 0 - 26 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.696 ms | 0 - 31 MB | NPU
| ResNet18 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.282 ms | 0 - 26 MB | NPU
| ResNet18 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.924 ms | 0 - 63 MB | NPU
| ResNet18 | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 5.96 ms | 0 - 27 MB | NPU
| ResNet18 | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 1.317 ms | 0 - 5 MB | NPU
| ResNet18 | TFLITE | float | Qualcomm® SA8775P | 1.968 ms | 0 - 31 MB | NPU
| ResNet18 | TFLITE | float | Qualcomm® QCS9075 | 2.03 ms | 0 - 25 MB | NPU
| ResNet18 | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 2.478 ms | 0 - 57 MB | NPU
| ResNet18 | TFLITE | float | Qualcomm® SA7255P | 5.96 ms | 0 - 27 MB | NPU
| ResNet18 | TFLITE | float | Qualcomm® SA8295P | 2.335 ms | 0 - 22 MB | NPU
| ResNet18 | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.735 ms | 0 - 25 MB | NPU
| ResNet18 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.588 ms | 0 - 30 MB | NPU
| ResNet18 | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.293 ms | 0 - 47 MB | NPU
| ResNet18 | TFLITE | w8a8 | Qualcomm® QCS6490 | 1.481 ms | 0 - 13 MB | NPU
| ResNet18 | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 1.041 ms | 0 - 23 MB | NPU
| ResNet18 | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.391 ms | 0 - 1 MB | NPU
| ResNet18 | TFLITE | w8a8 | Qualcomm® SA8775P | 0.555 ms | 0 - 25 MB | NPU
| ResNet18 | TFLITE | w8a8 | Qualcomm® QCS9075 | 0.464 ms | 0 - 13 MB | NPU
| ResNet18 | TFLITE | w8a8 | Qualcomm® QCM6690 | 2.788 ms | 0 - 30 MB | NPU
| ResNet18 | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.609 ms | 0 - 49 MB | NPU
| ResNet18 | TFLITE | w8a8 | Qualcomm® SA7255P | 1.041 ms | 0 - 23 MB | NPU
| ResNet18 | TFLITE | w8a8 | Qualcomm® SA8295P | 0.774 ms | 0 - 20 MB | NPU
| ResNet18 | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.246 ms | 0 - 21 MB | NPU
| ResNet18 | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.564 ms | 0 - 30 MB | NPU
| ResNet18 | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.233 ms | 0 - 25 MB | NPU

## License
* The license for the original implementation of ResNet18 can be found
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).

## References
* [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385)
* [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).