README.md

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

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/web-assets/model_demo.png)

# MobileNet-v2: Optimized for Qualcomm Devices

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 is based on the implementation of MobileNet-v2 found [here](https://github.com/tonylins/pytorch-mobilenet-v2/tree/master).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/mobilenet_v2) 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 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/releases/v0.48.0/mobilenet_v2-onnx-float.zip)
| ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/releases/v0.48.0/mobilenet_v2-onnx-w8a16.zip)
| ONNX | w8a16_mixed_int16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/releases/v0.48.0/mobilenet_v2-onnx-w8a16_mixed_int16.zip)
| ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/releases/v0.48.0/mobilenet_v2-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/releases/v0.48.0/mobilenet_v2-qnn_dlc-float.zip)
| QNN_DLC | w8a16 | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/releases/v0.48.0/mobilenet_v2-qnn_dlc-w8a16.zip)
| QNN_DLC | w8a16_mixed_int16 | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/releases/v0.48.0/mobilenet_v2-qnn_dlc-w8a16_mixed_int16.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/releases/v0.48.0/mobilenet_v2-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/releases/v0.48.0/mobilenet_v2-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v2/releases/v0.48.0/mobilenet_v2-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[MobileNet-v2 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/mobilenet_v2)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/mobilenet_v2) 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 [MobileNet-v2 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/mobilenet_v2) for usage instructions.

## 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

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| MobileNet-v2 | ONNX | float | Snapdragon® X2 Elite | 0.307 ms | 7 - 7 MB | NPU
| MobileNet-v2 | ONNX | float | Snapdragon® X Elite | 0.778 ms | 7 - 7 MB | NPU
| MobileNet-v2 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.428 ms | 0 - 52 MB | NPU
| MobileNet-v2 | ONNX | float | Qualcomm® QCS8550 (Proxy) | 0.62 ms | 1 - 2 MB | NPU
| MobileNet-v2 | ONNX | float | Qualcomm® QCS9075 | 0.893 ms | 1 - 3 MB | NPU
| MobileNet-v2 | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.338 ms | 0 - 34 MB | NPU
| MobileNet-v2 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.276 ms | 0 - 34 MB | NPU
| MobileNet-v2 | ONNX | w8a16 | Snapdragon® X2 Elite | 0.275 ms | 5 - 5 MB | NPU
| MobileNet-v2 | ONNX | w8a16 | Snapdragon® X Elite | 0.719 ms | 3 - 3 MB | NPU
| MobileNet-v2 | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.416 ms | 4 - 50 MB | NPU
| MobileNet-v2 | ONNX | w8a16 | Qualcomm® QCS6490 | 48.281 ms | 14 - 17 MB | CPU
| MobileNet-v2 | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 0.579 ms | 0 - 36 MB | NPU
| MobileNet-v2 | ONNX | w8a16 | Qualcomm® QCS9075 | 0.775 ms | 0 - 3 MB | NPU
| MobileNet-v2 | ONNX | w8a16 | Qualcomm® QCM6690 | 18.417 ms | 15 - 22 MB | CPU
| MobileNet-v2 | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.292 ms | 0 - 30 MB | NPU
| MobileNet-v2 | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 13.126 ms | 15 - 23 MB | CPU
| MobileNet-v2 | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.219 ms | 0 - 36 MB | NPU
| MobileNet-v2 | ONNX | w8a16_mixed_int16 | Snapdragon® X2 Elite | 0.306 ms | 6 - 6 MB | NPU
| MobileNet-v2 | ONNX | w8a16_mixed_int16 | Snapdragon® X Elite | 0.78 ms | 4 - 4 MB | NPU
| MobileNet-v2 | ONNX | w8a16_mixed_int16 | Snapdragon® 8 Gen 3 Mobile | 0.457 ms | 0 - 48 MB | NPU
| MobileNet-v2 | ONNX | w8a16_mixed_int16 | Qualcomm® QCS6490 | 46.897 ms | 15 - 17 MB | CPU
| MobileNet-v2 | ONNX | w8a16_mixed_int16 | Qualcomm® QCS8550 (Proxy) | 0.639 ms | 0 - 2 MB | NPU
| MobileNet-v2 | ONNX | w8a16_mixed_int16 | Qualcomm® QCS9075 | 0.812 ms | 0 - 3 MB | NPU
| MobileNet-v2 | ONNX | w8a16_mixed_int16 | Qualcomm® QCM6690 | 18.29 ms | 15 - 22 MB | CPU
| MobileNet-v2 | ONNX | w8a16_mixed_int16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.337 ms | 0 - 37 MB | NPU
| MobileNet-v2 | ONNX | w8a16_mixed_int16 | Snapdragon® 7 Gen 4 Mobile | 13.131 ms | 15 - 23 MB | CPU
| MobileNet-v2 | ONNX | w8a16_mixed_int16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.256 ms | 0 - 37 MB | NPU
| MobileNet-v2 | ONNX | w8a8 | Snapdragon® X2 Elite | 0.257 ms | 5 - 5 MB | NPU
| MobileNet-v2 | ONNX | w8a8 | Snapdragon® X Elite | 0.616 ms | 3 - 3 MB | NPU
| MobileNet-v2 | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.355 ms | 0 - 44 MB | NPU
| MobileNet-v2 | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.488 ms | 0 - 30 MB | NPU
| MobileNet-v2 | ONNX | w8a8 | Qualcomm® QCS9075 | 0.636 ms | 0 - 3 MB | NPU
| MobileNet-v2 | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.3 ms | 0 - 30 MB | NPU
| MobileNet-v2 | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.273 ms | 0 - 35 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Snapdragon® X2 Elite | 0.513 ms | 1 - 1 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Snapdragon® X Elite | 1.122 ms | 1 - 1 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.604 ms | 0 - 52 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 2.709 ms | 1 - 32 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 0.939 ms | 1 - 2 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Qualcomm® SA8775P | 1.253 ms | 1 - 33 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Qualcomm® QCS9075 | 1.127 ms | 1 - 3 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 1.729 ms | 0 - 55 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Qualcomm® SA7255P | 2.709 ms | 1 - 32 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Qualcomm® SA8295P | 1.522 ms | 0 - 29 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.46 ms | 1 - 31 MB | NPU
| MobileNet-v2 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.352 ms | 1 - 36 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 0.472 ms | 0 - 0 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Snapdragon® X Elite | 0.98 ms | 0 - 0 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.565 ms | 0 - 41 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Qualcomm® QCS6490 | 2.383 ms | 2 - 4 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 1.782 ms | 0 - 30 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 0.821 ms | 0 - 2 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Qualcomm® SA8775P | 1.032 ms | 0 - 30 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 0.987 ms | 0 - 2 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 3.364 ms | 0 - 141 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 0.983 ms | 0 - 44 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Qualcomm® SA7255P | 1.782 ms | 0 - 30 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Qualcomm® SA8295P | 1.295 ms | 1 - 28 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.383 ms | 0 - 28 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 0.866 ms | 0 - 141 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.298 ms | 0 - 32 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Snapdragon® X2 Elite | 0.47 ms | 0 - 0 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Snapdragon® X Elite | 1.013 ms | 0 - 0 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 8 Gen 3 Mobile | 0.608 ms | 0 - 41 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCS8275 (Proxy) | 1.985 ms | 0 - 30 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCS8550 (Proxy) | 0.867 ms | 0 - 2 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Qualcomm® SA8775P | 1.08 ms | 0 - 32 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCS9075 | 1.033 ms | 0 - 2 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Qualcomm® QCM6690 | 4.775 ms | 0 - 142 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Qualcomm® SA7255P | 1.985 ms | 0 - 30 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.414 ms | 0 - 29 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 7 Gen 4 Mobile | 0.966 ms | 0 - 142 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.324 ms | 0 - 33 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.288 ms | 0 - 0 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.557 ms | 0 - 0 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.314 ms | 0 - 40 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 1.372 ms | 2 - 4 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 1.035 ms | 0 - 30 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.442 ms | 0 - 2 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.611 ms | 0 - 30 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 0.533 ms | 0 - 2 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 1.764 ms | 0 - 28 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.6 ms | 0 - 41 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Qualcomm® SA7255P | 1.035 ms | 0 - 30 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Qualcomm® SA8295P | 0.818 ms | 0 - 27 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.215 ms | 0 - 32 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.47 ms | 0 - 28 MB | NPU
| MobileNet-v2 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.181 ms | 0 - 31 MB | NPU
| MobileNet-v2 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.598 ms | 0 - 56 MB | NPU
| MobileNet-v2 | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 2.757 ms | 0 - 34 MB | NPU
| MobileNet-v2 | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 0.924 ms | 0 - 2 MB | NPU
| MobileNet-v2 | TFLITE | float | Qualcomm® SA8775P | 1.267 ms | 0 - 38 MB | NPU
| MobileNet-v2 | TFLITE | float | Qualcomm® QCS9075 | 1.13 ms | 0 - 10 MB | NPU
| MobileNet-v2 | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 1.723 ms | 0 - 57 MB | NPU
| MobileNet-v2 | TFLITE | float | Qualcomm® SA7255P | 2.757 ms | 0 - 34 MB | NPU
| MobileNet-v2 | TFLITE | float | Qualcomm® SA8295P | 1.519 ms | 0 - 32 MB | NPU
| MobileNet-v2 | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.45 ms | 0 - 35 MB | NPU
| MobileNet-v2 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.351 ms | 0 - 39 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.288 ms | 0 - 41 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Qualcomm® QCS6490 | 1.208 ms | 0 - 7 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 0.981 ms | 0 - 30 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.417 ms | 0 - 2 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Qualcomm® SA8775P | 2.372 ms | 0 - 30 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Qualcomm® QCS9075 | 0.55 ms | 0 - 6 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Qualcomm® QCM6690 | 1.641 ms | 0 - 29 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.547 ms | 0 - 45 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Qualcomm® SA7255P | 0.981 ms | 0 - 30 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Qualcomm® SA8295P | 0.807 ms | 0 - 27 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.215 ms | 0 - 33 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.421 ms | 0 - 28 MB | NPU
| MobileNet-v2 | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.183 ms | 0 - 35 MB | NPU

## License
* The license for the original implementation of MobileNet-v2 can be found
  [here](https://github.com/tonylins/pytorch-mobilenet-v2/blob/master/LICENSE).

## References
* [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381)
* [Source Model Implementation](https://github.com/tonylins/pytorch-mobilenet-v2/tree/master)

## 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).