README.md

---
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/efficientnet_lite4/web-assets/model_demo.png)

# EfficientNet-Lite4: Optimized for Qualcomm Devices

EfficientNet-Lite4 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 EfficientNet-Lite4 found [here](https://github.com/RangiLyu/EfficientNet-Lite).
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/v0.58.0/src/qai_hub_models/models/efficientnet_lite4) 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 2.45, ONNX Runtime 1.26.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.45, ONNX Runtime 1.26.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[EfficientNet-Lite4 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/efficientnet_lite4)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/efficientnet_lite4) 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 [EfficientNet-Lite4 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/efficientnet_lite4) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Model checkpoint: Imagenet
- Input resolution: 300x300
- Number of parameters: 13.01M
- Model size (float): 51.9 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| EfficientNet-Lite4 | ONNX | float | Snapdragon® X2 Elite | 1.442 ms | 1 - 1 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Snapdragon® X Elite | 2.735 ms | 29 - 29 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 1.912 ms | 0 - 102 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 5.53 ms | 1 - 100 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2.633 ms | 0 - 48 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Qualcomm® QCS8450 | 5.53 ms | 1 - 100 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 4.127 ms | 1 - 4 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Snapdragon® 8 Elite Mobile | 1.497 ms | 0 - 55 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.145 ms | 0 - 57 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 1.497 ms | 0 - 55 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 2.735 ms | 29 - 29 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® X2 Elite | 0.648 ms | 1 - 1 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® X Elite | 1.269 ms | 16 - 16 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.877 ms | 0 - 109 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 2.073 ms | 0 - 117 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 6.201 ms | 0 - 3 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.227 ms | 0 - 3 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® QCS8450 | 2.073 ms | 0 - 117 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.604 ms | 0 - 75 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.581 ms | 0 - 73 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 10.908 ms | 0 - 192 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 0.719 ms | 0 - 74 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.414 ms | 0 - 3 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 1.581 ms | 0 - 73 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.719 ms | 0 - 74 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 1.269 ms | 16 - 16 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® X2 Elite | 1.864 ms | 1 - 1 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® X Elite | 3.534 ms | 1 - 1 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 2.19 ms | 0 - 103 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 6.909 ms | 0 - 98 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® QCS8275 | 15.285 ms | 1 - 53 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.166 ms | 1 - 3 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® SA8775P | 4.79 ms | 1 - 55 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® SA8650P | 4.79 ms | 1 - 55 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® SA8255P | 4.79 ms | 1 - 55 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® QCS8450 | 6.909 ms | 0 - 98 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® SA8295P | 6.439 ms | 1 - 50 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 4.511 ms | 3 - 6 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 1.687 ms | 1 - 55 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® SA7255P | 15.285 ms | 1 - 53 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.244 ms | 1 - 57 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 1.687 ms | 1 - 55 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 3.534 ms | 1 - 1 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.722 ms | 0 - 0 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® X Elite | 1.448 ms | 0 - 0 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.914 ms | 0 - 94 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 2.349 ms | 0 - 105 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 7.272 ms | 2 - 4 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® QCS8275 | 3.327 ms | 0 - 64 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.249 ms | 0 - 2 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 1.613 ms | 0 - 64 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® SA8650P | 1.613 ms | 0 - 64 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® SA8255P | 1.613 ms | 0 - 64 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 2.349 ms | 0 - 105 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.529 ms | 0 - 67 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.623 ms | 0 - 65 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® SA8295P | 2.53 ms | 0 - 61 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 11.23 ms | 0 - 185 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 0.653 ms | 0 - 62 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.431 ms | 0 - 2 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® SA7255P | 3.327 ms | 0 - 64 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 1.623 ms | 0 - 65 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.653 ms | 0 - 62 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 1.448 ms | 0 - 0 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 2.193 ms | 0 - 114 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 6.671 ms | 0 - 112 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® QCS8275 | 15.283 ms | 0 - 65 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.113 ms | 0 - 3 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® SA8775P | 4.801 ms | 0 - 66 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® SA8650P | 4.801 ms | 0 - 66 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® SA8255P | 4.801 ms | 0 - 66 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® QCS8450 | 6.671 ms | 0 - 112 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® SA8295P | 6.427 ms | 0 - 48 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 4.527 ms | 0 - 32 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Snapdragon® 8 Elite Mobile | 1.686 ms | 0 - 66 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® SA7255P | 15.283 ms | 0 - 65 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.258 ms | 0 - 68 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 1.686 ms | 0 - 66 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.66 ms | 0 - 98 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 1.739 ms | 0 - 104 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 5.834 ms | 0 - 31 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® QCS8275 | 2.815 ms | 0 - 59 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.914 ms | 0 - 15 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® SA8775P | 1.268 ms | 0 - 63 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® SA8650P | 1.268 ms | 0 - 63 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® SA8255P | 1.268 ms | 0 - 63 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® QCS8450 | 1.739 ms | 0 - 104 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.419 ms | 0 - 63 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.244 ms | 0 - 174 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® SA8295P | 2.145 ms | 0 - 56 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 10.036 ms | 0 - 178 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 0.518 ms | 0 - 57 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.089 ms | 0 - 17 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® SA7255P | 2.815 ms | 0 - 59 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 1.244 ms | 0 - 174 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.518 ms | 0 - 57 MB | NPU

## License
* The license for the original implementation of EfficientNet-Lite4 can be found
  [here](https://github.com/RangiLyu/EfficientNet-Lite/blob/main/LICENSE).

## References
* [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://arxiv.org/abs/1905.11946)
* [Source Model Implementation](https://github.com/RangiLyu/EfficientNet-Lite)

## 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).