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_b0/web-assets/model_demo.png)

# EfficientNet-B0: Optimized for Qualcomm Devices

EfficientNetB0 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-B0 found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/efficientnet.py).
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/v0.59.0/src/qai_hub_models/models/efficientnet_b0) 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 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_b0/releases/v0.59.0/efficientnet_b0-onnx-float.zip)
| ONNX | w8a16 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_b0/releases/v0.59.0/efficientnet_b0-onnx-w8a16.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_b0/releases/v0.59.0/efficientnet_b0-qnn_dlc-float.zip)
| QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_b0/releases/v0.59.0/efficientnet_b0-qnn_dlc-w8a16.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_b0/releases/v0.59.0/efficientnet_b0-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[EfficientNet-B0 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/efficientnet_b0)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/efficientnet_b0) 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-B0 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/efficientnet_b0) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Model checkpoint: Imagenet
- Input resolution: 224x224
- Number of parameters: 5.27M
- Model size (float): 20.1 MB
- Model size (w8a16): 6.99 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| EfficientNet-B0 | ONNX | float | Snapdragon® X2 Elite | 0.652 ms | 2 - 2 MB | NPU
| EfficientNet-B0 | ONNX | float | Snapdragon® X Elite | 1.306 ms | 14 - 14 MB | NPU
| EfficientNet-B0 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.898 ms | 0 - 65 MB | NPU
| EfficientNet-B0 | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 2.873 ms | 1 - 77 MB | NPU
| EfficientNet-B0 | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.28 ms | 0 - 17 MB | NPU
| EfficientNet-B0 | ONNX | float | Qualcomm® QCS8450 | 2.873 ms | 1 - 77 MB | NPU
| EfficientNet-B0 | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 1.622 ms | 1 - 3 MB | NPU
| EfficientNet-B0 | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 1.306 ms | 14 - 14 MB | NPU
| EfficientNet-B0 | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 0.697 ms | 0 - 42 MB | NPU
| EfficientNet-B0 | ONNX | float | Snapdragon® 8 Elite Mobile | 0.697 ms | 0 - 42 MB | NPU
| EfficientNet-B0 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.546 ms | 0 - 43 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Snapdragon® X2 Elite | 0.567 ms | 1 - 1 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Snapdragon® X Elite | 1.485 ms | 6 - 6 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.939 ms | 0 - 83 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 1.663 ms | 0 - 83 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 3.737 ms | 0 - 3 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.422 ms | 0 - 10 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Qualcomm® QCS8450 | 1.663 ms | 0 - 83 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 1.789 ms | 0 - 3 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 1.485 ms | 6 - 6 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 0.668 ms | 0 - 64 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 0.668 ms | 0 - 64 MB | NPU
| EfficientNet-B0 | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.554 ms | 0 - 63 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Snapdragon® X2 Elite | 0.84 ms | 1 - 1 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Snapdragon® X Elite | 1.783 ms | 1 - 1 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 1.074 ms | 0 - 64 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 3.593 ms | 0 - 73 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 4.925 ms | 1 - 37 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.558 ms | 1 - 2 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® SA8775P | 2.012 ms | 1 - 40 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® SA8650P | 2.012 ms | 1 - 40 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® SA8255P | 2.012 ms | 1 - 40 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® QCS8450 | 3.593 ms | 0 - 73 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 2.338 ms | 1 - 3 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 1.783 ms | 1 - 1 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 0.805 ms | 1 - 37 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® SA7255P | 4.925 ms | 1 - 37 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Qualcomm® SA8295P | 3.615 ms | 0 - 44 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 0.805 ms | 1 - 37 MB | NPU
| EfficientNet-B0 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.611 ms | 1 - 38 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 0.795 ms | 0 - 0 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Snapdragon® X Elite | 1.862 ms | 0 - 0 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.132 ms | 0 - 65 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Snapdragon® 8 Gen 1 Mobile | 1.973 ms | 0 - 70 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 5.122 ms | 0 - 2 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8275 | 3.303 ms | 0 - 48 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.676 ms | 0 - 43 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® SA8775P | 1.951 ms | 0 - 51 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® SA8650P | 1.951 ms | 0 - 51 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® SA8255P | 1.951 ms | 0 - 51 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® QCS8450 | 1.973 ms | 0 - 70 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 1.783 ms | 2 - 4 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 1.862 ms | 0 - 0 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 6.526 ms | 0 - 168 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 1.712 ms | 0 - 48 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 0.776 ms | 0 - 52 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® SA7255P | 3.303 ms | 0 - 48 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Qualcomm® SA8295P | 2.364 ms | 0 - 46 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 0.776 ms | 0 - 52 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.644 ms | 0 - 52 MB | NPU
| EfficientNet-B0 | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1.712 ms | 0 - 48 MB | NPU
| EfficientNet-B0 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.071 ms | 0 - 66 MB | NPU
| EfficientNet-B0 | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 3.605 ms | 0 - 81 MB | NPU
| EfficientNet-B0 | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 4.931 ms | 0 - 41 MB | NPU
| EfficientNet-B0 | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.533 ms | 0 - 2 MB | NPU
| EfficientNet-B0 | TFLITE | float | Qualcomm® SA8775P | 2.05 ms | 0 - 45 MB | NPU
| EfficientNet-B0 | TFLITE | float | Qualcomm® SA8650P | 2.05 ms | 0 - 45 MB | NPU
| EfficientNet-B0 | TFLITE | float | Qualcomm® SA8255P | 2.05 ms | 0 - 45 MB | NPU
| EfficientNet-B0 | TFLITE | float | Qualcomm® QCS8450 | 3.605 ms | 0 - 81 MB | NPU
| EfficientNet-B0 | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 1.887 ms | 0 - 16 MB | NPU
| EfficientNet-B0 | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 0.815 ms | 0 - 41 MB | NPU
| EfficientNet-B0 | TFLITE | float | Qualcomm® SA7255P | 4.931 ms | 0 - 41 MB | NPU
| EfficientNet-B0 | TFLITE | float | Qualcomm® SA8295P | 3.613 ms | 0 - 48 MB | NPU
| EfficientNet-B0 | TFLITE | float | Snapdragon® 8 Elite Mobile | 0.815 ms | 0 - 41 MB | NPU
| EfficientNet-B0 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.612 ms | 0 - 42 MB | NPU

## License
* The license for the original implementation of EfficientNet-B0 can be found
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).

## References
* [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://arxiv.org/abs/1905.11946)
* [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/efficientnet.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).