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_b4/web-assets/model_demo.png)

# EfficientNet-B4: Optimized for Qualcomm Devices

EfficientNetB4 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-B4 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_b4) 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_b4/releases/v0.59.0/efficientnet_b4-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_b4/releases/v0.59.0/efficientnet_b4-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_b4/releases/v0.59.0/efficientnet_b4-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_b4/releases/v0.59.0/efficientnet_b4-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_b4/releases/v0.59.0/efficientnet_b4-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[EfficientNet-B4 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/efficientnet_b4)**.


### 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_b4) 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-B4 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/efficientnet_b4) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Model checkpoint: Imagenet
- Input resolution: 380x380
- Number of parameters: 19.3M
- Model size (float): 73.6 MB
- Model size (w8a16): 24.0 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| EfficientNet-B4 | ONNX | float | Snapdragon® X2 Elite | 3.918 ms | 2 - 2 MB | NPU
| EfficientNet-B4 | ONNX | float | Snapdragon® X Elite | 7.724 ms | 45 - 45 MB | NPU
| EfficientNet-B4 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 5.31 ms | 2 - 149 MB | NPU
| EfficientNet-B4 | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 20.504 ms | 0 - 187 MB | NPU
| EfficientNet-B4 | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 7.341 ms | 0 - 218 MB | NPU
| EfficientNet-B4 | ONNX | float | Qualcomm® QCS8450 | 20.504 ms | 0 - 187 MB | NPU
| EfficientNet-B4 | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 9.342 ms | 1 - 6 MB | NPU
| EfficientNet-B4 | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 7.724 ms | 45 - 45 MB | NPU
| EfficientNet-B4 | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 4.052 ms | 0 - 89 MB | NPU
| EfficientNet-B4 | ONNX | float | Snapdragon® 8 Elite Mobile | 4.052 ms | 0 - 89 MB | NPU
| EfficientNet-B4 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.113 ms | 0 - 205 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Snapdragon® X2 Elite | 2.917 ms | 2 - 2 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Snapdragon® X Elite | 8.032 ms | 23 - 23 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 5.079 ms | 1 - 226 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 8.907 ms | 0 - 227 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 33.142 ms | 0 - 4 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 7.615 ms | 0 - 30 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Qualcomm® QCS8450 | 8.907 ms | 0 - 227 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 7.96 ms | 1 - 4 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 8.032 ms | 23 - 23 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 3.413 ms | 0 - 170 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 3.413 ms | 0 - 170 MB | NPU
| EfficientNet-B4 | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 2.75 ms | 0 - 178 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Snapdragon® X2 Elite | 4.558 ms | 2 - 2 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Snapdragon® X Elite | 8.862 ms | 2 - 2 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 5.807 ms | 0 - 142 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 23.025 ms | 2 - 187 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 29.127 ms | 2 - 82 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 8.151 ms | 2 - 165 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® SA8775P | 10.296 ms | 2 - 85 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® SA8650P | 10.296 ms | 2 - 85 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® SA8255P | 10.296 ms | 2 - 85 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® QCS8450 | 23.025 ms | 2 - 187 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 10.032 ms | 4 - 7 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 8.862 ms | 2 - 2 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 4.302 ms | 2 - 89 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® SA7255P | 29.127 ms | 2 - 82 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Qualcomm® SA8295P | 18.782 ms | 2 - 123 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 4.302 ms | 2 - 89 MB | NPU
| EfficientNet-B4 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.356 ms | 2 - 208 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 3.553 ms | 1 - 1 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Snapdragon® X Elite | 9.1 ms | 1 - 1 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 5.63 ms | 1 - 197 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Snapdragon® 8 Gen 1 Mobile | 11.375 ms | 1 - 201 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 33.639 ms | 3 - 5 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8275 | 15.405 ms | 1 - 141 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 8.363 ms | 1 - 212 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® SA8775P | 8.902 ms | 1 - 143 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® SA8650P | 8.902 ms | 1 - 143 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® SA8255P | 8.902 ms | 1 - 143 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® QCS8450 | 11.375 ms | 1 - 201 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 8.648 ms | 1 - 3 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 9.1 ms | 1 - 1 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 48.82 ms | 1 - 275 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 9.736 ms | 1 - 269 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 3.704 ms | 0 - 145 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® SA7255P | 15.405 ms | 1 - 141 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Qualcomm® SA8295P | 10.916 ms | 1 - 143 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 3.704 ms | 0 - 145 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 3.012 ms | 1 - 156 MB | NPU
| EfficientNet-B4 | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 9.736 ms | 1 - 269 MB | NPU
| EfficientNet-B4 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 5.754 ms | 0 - 163 MB | NPU
| EfficientNet-B4 | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 21.764 ms | 0 - 203 MB | NPU
| EfficientNet-B4 | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 28.922 ms | 0 - 97 MB | NPU
| EfficientNet-B4 | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 8.016 ms | 0 - 2 MB | NPU
| EfficientNet-B4 | TFLITE | float | Qualcomm® SA8775P | 10.288 ms | 0 - 100 MB | NPU
| EfficientNet-B4 | TFLITE | float | Qualcomm® SA8650P | 10.288 ms | 0 - 100 MB | NPU
| EfficientNet-B4 | TFLITE | float | Qualcomm® SA8255P | 10.288 ms | 0 - 100 MB | NPU
| EfficientNet-B4 | TFLITE | float | Qualcomm® QCS8450 | 21.764 ms | 0 - 203 MB | NPU
| EfficientNet-B4 | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 9.989 ms | 0 - 49 MB | NPU
| EfficientNet-B4 | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 4.317 ms | 0 - 104 MB | NPU
| EfficientNet-B4 | TFLITE | float | Qualcomm® SA7255P | 28.922 ms | 0 - 97 MB | NPU
| EfficientNet-B4 | TFLITE | float | Qualcomm® SA8295P | 18.851 ms | 0 - 139 MB | NPU
| EfficientNet-B4 | TFLITE | float | Snapdragon® 8 Elite Mobile | 4.317 ms | 0 - 104 MB | NPU
| EfficientNet-B4 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.216 ms | 0 - 98 MB | NPU

## License
* The license for the original implementation of EfficientNet-B4 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).