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

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/web-assets/model_demo.png)

# MobileNet-v4: Optimized for Qualcomm Devices

MobileNetV4 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-v4 found [here](https://github.com/jaiwei98/MobileNetV4-pytorch).
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/v0.66.0/src/qai_hub_models/models/mobilenet_v4) 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 2.50, ONNX Runtime 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.50, ONNX Runtime 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[MobileNet-v4 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/mobilenet_v4)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.66.0/src/qai_hub_models/models/mobilenet_v4) 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-v4 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.66.0/src/qai_hub_models/models/mobilenet_v4) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Input resolution: 224x224
- Model checkpoint: Imagenet
- Model size (float): 37.0 MB
- Model size (w8a8): 10.2 MB
- Number of parameters: 9.68M

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| MobileNet-v4 | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.444 ms | 0 - 43 MB | NPU
| MobileNet-v4 | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.503 ms | 0 - 39 MB | NPU
| MobileNet-v4 | ONNX | float | Snapdragon® X2 Elite | 0.379 ms | 2 - 2 MB | NPU
| MobileNet-v4 | ONNX | float | Snapdragon® X Elite | 0.823 ms | 22 - 22 MB | NPU
| MobileNet-v4 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.613 ms | 0 - 74 MB | NPU
| MobileNet-v4 | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 1.436 ms | 0 - 77 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 1.27 ms | 1 - 5 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.865 ms | 0 - 37 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® QCS8450 | 1.436 ms | 0 - 77 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 1.201 ms | 1 - 4 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 0.823 ms | 22 - 22 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 0.503 ms | 0 - 39 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.337 ms | 0 - 53 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.367 ms | 0 - 53 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® X2 Elite | 0.302 ms | 1 - 1 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® X Elite | 0.548 ms | 10 - 10 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.426 ms | 0 - 84 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 0.778 ms | 0 - 89 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 1.733 ms | 0 - 3 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 0.685 ms | 0 - 4 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.578 ms | 0 - 76 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® QCS8450 | 0.778 ms | 0 - 89 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 0.691 ms | 0 - 3 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 0.548 ms | 10 - 10 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 2.585 ms | 0 - 54 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 0.616 ms | 0 - 54 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.367 ms | 0 - 53 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.616 ms | 0 - 54 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.526 ms | 1 - 40 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.625 ms | 1 - 38 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® X2 Elite | 0.582 ms | 1 - 1 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® X Elite | 1.301 ms | 1 - 1 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.82 ms | 0 - 68 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 2.128 ms | 0 - 77 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 1.486 ms | 1 - 4 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.18 ms | 1 - 2 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® SA8775P | 1.589 ms | 1 - 42 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® SA8650P | 1.589 ms | 1 - 42 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® SA8255P | 1.589 ms | 1 - 42 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® QCS8450 | 2.128 ms | 0 - 77 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 1.468 ms | 1 - 3 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 1.301 ms | 1 - 1 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 0.625 ms | 1 - 38 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® SA8295P | 2.05 ms | 0 - 38 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.276 ms | 0 - 49 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.309 ms | 0 - 51 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.347 ms | 0 - 0 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.659 ms | 0 - 0 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.419 ms | 0 - 72 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 0.77 ms | 0 - 76 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 2.097 ms | 2 - 4 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 0.634 ms | 0 - 3 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.563 ms | 0 - 2 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.739 ms | 0 - 52 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® SA8650P | 0.739 ms | 0 - 52 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® SA8255P | 0.739 ms | 0 - 52 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 0.77 ms | 0 - 76 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 0.648 ms | 0 - 2 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 0.659 ms | 0 - 0 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 2.931 ms | 0 - 49 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 0.698 ms | 0 - 50 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.309 ms | 0 - 51 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® SA7255P | 1.305 ms | 0 - 48 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® SA8295P | 0.998 ms | 0 - 46 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.698 ms | 0 - 50 MB | NPU
| MobileNet-v4 | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.517 ms | 0 - 52 MB | NPU
| MobileNet-v4 | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.624 ms | 0 - 52 MB | NPU
| MobileNet-v4 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.802 ms | 0 - 85 MB | NPU
| MobileNet-v4 | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 2.137 ms | 0 - 95 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 1.476 ms | 0 - 25 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.14 ms | 0 - 2 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® SA8775P | 1.597 ms | 0 - 55 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® SA8650P | 1.597 ms | 0 - 55 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® SA8255P | 1.597 ms | 0 - 55 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® QCS8450 | 2.137 ms | 0 - 95 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 1.479 ms | 0 - 25 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 0.624 ms | 0 - 52 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® SA7255P | 4.314 ms | 0 - 52 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® SA8295P | 2.067 ms | 0 - 51 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.219 ms | 0 - 46 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.241 ms | 0 - 44 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.297 ms | 0 - 73 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 0.578 ms | 0 - 81 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 1.649 ms | 0 - 19 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 0.485 ms | 0 - 14 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.389 ms | 0 - 5 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® SA8775P | 0.576 ms | 0 - 50 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® SA8650P | 0.576 ms | 0 - 50 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® SA8255P | 0.576 ms | 0 - 50 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® QCS8450 | 0.578 ms | 0 - 81 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 0.489 ms | 0 - 13 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 2.332 ms | 0 - 46 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 0.529 ms | 0 - 46 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.241 ms | 0 - 44 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® SA7255P | 0.981 ms | 0 - 46 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® SA8295P | 0.806 ms | 0 - 43 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.529 ms | 0 - 46 MB | NPU

## License
* The license for the original implementation of MobileNet-v4 can be found
  [here](https://github.com/jaiwei98/MobileNetV4-pytorch/blob/main/LICENSE).

## References
* [MobileNetV4 -- Universal Models for the Mobile Ecosystem](https://arxiv.org/abs/2404.10518)
* [Source Model Implementation](https://github.com/jaiwei98/MobileNetV4-pytorch)

## 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).