# EfficientNet Keras (and TensorFlow Keras)
[](https://badge.fury.io/py/efficientnet) [](https://pepy.tech/project/efficientnet/month)
This repository contains a Keras (and TensorFlow Keras) reimplementation of **EfficientNet**, a lightweight convolutional neural network architecture achieving the [state-of-the-art accuracy with an order of magnitude fewer parameters and FLOPS](https://arxiv.org/abs/1905.11946), on both ImageNet and
five other commonly used transfer learning datasets.
The codebase is heavily inspired by the [TensorFlow implementation](https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet).
## Important!
There was a huge library update **24 of July 2019**. Now efficintnet works with both frameworks: `keras` and `tensorflow.keras`.
If you have models, trained before that date, to load them, please, use efficientnet of 0.0.4 version (PyPI). You can roll back using `pip install -U efficientnet==0.0.4`.
## Table of Contents
1. [About EfficientNet Models](#about-efficientnet-models)
2. [Examples](#examples)
3. [Models](#models)
4. [Installation](#installation)
5. [Frequently Asked Questions](#frequently-asked-questions)
6. [Acknowledgements](#acknowledgements)
## About EfficientNet Models
EfficientNets rely on AutoML and compound scaling to achieve superior performance without compromising resource efficiency. The [AutoML Mobile framework](https://ai.googleblog.com/2018/08/mnasnet-towards-automating-design-of.html) has helped develop a mobile-size baseline network, **EfficientNet-B0**, which is then improved by the compound scaling method to obtain EfficientNet-B1 to B7.
<table border="0">
<tr>
<td>
<img src="https://raw.githubusercontent.com/tensorflow/tpu/master/models/official/efficientnet/g3doc/params.png" width="100%" />
</td>
<td>
<img src="https://raw.githubusercontent.com/tensorflow/tpu/master/models/official/efficientnet/g3doc/flops.png", width="90%" />
</td>
</tr>
</table>
EfficientNets achieve state-of-the-art accuracy on ImageNet with an order of magnitude better efficiency:
* In high-accuracy regime, EfficientNet-B7 achieves the state-of-the-art 84.4% top-1 / 97.1% top-5 accuracy on ImageNet with 66M parameters and 37B FLOPS. At the same time, the model is 8.4x smaller and 6.1x faster on CPU inference than the former leader, [Gpipe](https://arxiv.org/abs/1811.06965).
* In middle-accuracy regime, EfficientNet-B1 is 7.6x smaller and 5.7x faster on CPU inference than [ResNet-152](https://arxiv.org/abs/1512.03385), with similar ImageNet accuracy.
* Compared to the widely used [ResNet-50](https://arxiv.org/abs/1512.03385), EfficientNet-B4 improves the top-1 accuracy from 76.3% of ResNet-50 to 82.6% (+6.3%), under similar FLOPS constraints.
## Examples
* *Initializing the model*:
```python
# models can be build with Keras or Tensorflow frameworks
# use keras and tfkeras modules respectively
# efficientnet.keras / efficientnet.tfkeras
import efficientnet.keras as efn
model = efn.EfficientNetB0(weights='imagenet') # or weights='noisy-student'
```
* *Loading the pre-trained weights*:
```python
# model use some custom objects, so before loading saved model
# import module your network was build with
# e.g. import efficientnet.keras / import efficientnet.tfkeras
import efficientnet.tfkeras
from tensorflow.keras.models import load_model
model = load_model('path/to/model.h5')
```
See the complete example of loading the model and making an inference in the Jupyter notebook [here](https://github.com/qubvel/efficientnet/blob/master/examples/inference_example.ipynb).
## Models
The performance of each model variant using the pre-trained weights converted from checkpoints provided by the authors is as follows:
| Architecture | @top1* Imagenet| @top1* Noisy-Student|
| -------------- | :----: |:---:|
| EfficientNetB0 | 0.772 |0.788|
| EfficientNetB1 | 0.791 |0.815|
| EfficientNetB2 | 0.802 |0.824|
| EfficientNetB3 | 0.816 |0.841|
| EfficientNetB4 | 0.830 |0.853|
| EfficientNetB5 | 0.837 |0.861|
| EfficientNetB6 | 0.841 |0.864|
| EfficientNetB7 | 0.844 |0.869|
**\*** - topK accuracy score for converted models (imagenet `val` set)
## Installation
### Requirements
* `Keras >= 2.2.0` / `TensorFlow >= 1.12.0`
* `keras_applications >= 1.0.7`
* `scikit-image`
### Installing from the source
```bash
$ pip install -U git+https://github.com/qubvel/efficientnet
```
### Installing from PyPI
PyPI stable release
```bash
$ pip install -U efficientnet
```
PyPI latest release (with keras and tf.keras support)
```bash
$ pip install -U --pre efficientnet
```
## Frequently Asked Questions
* **How can I convert the original TensorFlow checkpoints to Keras HDF5?**
Pick the target directory (like `dist`) and run the [converter script](./scripts) from the repo directory as follows:
```bash
$ ./scripts/convert_efficientnet.sh --target_dir dist
```
You can also optionally create the virtual environment with all the dependencies installed by adding `--make_venv=true` and operate in a self-destructing temporary location instead of the target directory by setting `--tmp_working_dir=true`.
## Acknowledgements
I would like to thanks community members who actively contribute to this repository:
1) Sasha Illarionov ([@sdll](https://github.com/sdll)) for preparing automated script for weights conversion
2) Björn Barz ([@Callidior](https://github.com/Callidior)) for model code adaptation for keras and tensorflow.keras frameworks
Raw data
{
"_id": null,
"home_page": "https://github.com/qubvel/efficientnet",
"name": "efficientnet",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.0.0",
"maintainer_email": "",
"keywords": "",
"author": "Pavel Yakubovskiy",
"author_email": "qubvel@gmail.com",
"download_url": "https://files.pythonhosted.org/packages/bc/0f/811c73e9e579361b202b1e8205fff114ee7f9a738489247207c9141266f3/efficientnet-1.1.1.tar.gz",
"platform": "",
"description": "\n# EfficientNet Keras (and TensorFlow Keras)\n\n[](https://badge.fury.io/py/efficientnet) [](https://pepy.tech/project/efficientnet/month)\n\nThis repository contains a Keras (and TensorFlow Keras) reimplementation of **EfficientNet**, a lightweight convolutional neural network architecture achieving the [state-of-the-art accuracy with an order of magnitude fewer parameters and FLOPS](https://arxiv.org/abs/1905.11946), on both ImageNet and\nfive other commonly used transfer learning datasets.\n\nThe codebase is heavily inspired by the [TensorFlow implementation](https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet).\n\n## Important!\nThere was a huge library update **24 of July 2019**. Now efficintnet works with both frameworks: `keras` and `tensorflow.keras`.\nIf you have models, trained before that date, to load them, please, use efficientnet of 0.0.4 version (PyPI). You can roll back using `pip install -U efficientnet==0.0.4`.\n\n## Table of Contents\n\n 1. [About EfficientNet Models](#about-efficientnet-models)\n 2. [Examples](#examples)\n 3. [Models](#models)\n 4. [Installation](#installation)\n 5. [Frequently Asked Questions](#frequently-asked-questions)\n 6. [Acknowledgements](#acknowledgements)\n\n## About EfficientNet Models\n\nEfficientNets rely on AutoML and compound scaling to achieve superior performance without compromising resource efficiency. The [AutoML Mobile framework](https://ai.googleblog.com/2018/08/mnasnet-towards-automating-design-of.html) has helped develop a mobile-size baseline network, **EfficientNet-B0**, which is then improved by the compound scaling method to obtain EfficientNet-B1 to B7.\n\n<table border=\"0\">\n<tr>\n <td>\n <img src=\"https://raw.githubusercontent.com/tensorflow/tpu/master/models/official/efficientnet/g3doc/params.png\" width=\"100%\" />\n </td>\n <td>\n <img src=\"https://raw.githubusercontent.com/tensorflow/tpu/master/models/official/efficientnet/g3doc/flops.png\", width=\"90%\" />\n </td>\n</tr>\n</table>\n\nEfficientNets achieve state-of-the-art accuracy on ImageNet with an order of magnitude better efficiency:\n\n* In high-accuracy regime, EfficientNet-B7 achieves the state-of-the-art 84.4% top-1 / 97.1% top-5 accuracy on ImageNet with 66M parameters and 37B FLOPS. At the same time, the model is 8.4x smaller and 6.1x faster on CPU inference than the former leader, [Gpipe](https://arxiv.org/abs/1811.06965).\n\n* In middle-accuracy regime, EfficientNet-B1 is 7.6x smaller and 5.7x faster on CPU inference than [ResNet-152](https://arxiv.org/abs/1512.03385), with similar ImageNet accuracy.\n\n* Compared to the widely used [ResNet-50](https://arxiv.org/abs/1512.03385), EfficientNet-B4 improves the top-1 accuracy from 76.3% of ResNet-50 to 82.6% (+6.3%), under similar FLOPS constraints.\n\n## Examples\n\n* *Initializing the model*:\n\n```python\n# models can be build with Keras or Tensorflow frameworks\n# use keras and tfkeras modules respectively\n# efficientnet.keras / efficientnet.tfkeras\nimport efficientnet.keras as efn \n\nmodel = efn.EfficientNetB0(weights='imagenet') # or weights='noisy-student'\n\n```\n\n* *Loading the pre-trained weights*:\n\n```python\n# model use some custom objects, so before loading saved model\n# import module your network was build with\n# e.g. import efficientnet.keras / import efficientnet.tfkeras\nimport efficientnet.tfkeras\nfrom tensorflow.keras.models import load_model\n\nmodel = load_model('path/to/model.h5')\n```\n\nSee the complete example of loading the model and making an inference in the Jupyter notebook [here](https://github.com/qubvel/efficientnet/blob/master/examples/inference_example.ipynb).\n\n## Models\n\nThe performance of each model variant using the pre-trained weights converted from checkpoints provided by the authors is as follows:\n\n| Architecture | @top1* Imagenet| @top1* Noisy-Student| \n| -------------- | :----: |:---:|\n| EfficientNetB0 | 0.772 |0.788|\n| EfficientNetB1 | 0.791 |0.815|\n| EfficientNetB2 | 0.802 |0.824|\n| EfficientNetB3 | 0.816 |0.841|\n| EfficientNetB4 | 0.830 |0.853|\n| EfficientNetB5 | 0.837 |0.861|\n| EfficientNetB6 | 0.841 |0.864|\n| EfficientNetB7 | 0.844 |0.869|\n\n**\\*** - topK accuracy score for converted models (imagenet `val` set)\n\n## Installation\n\n### Requirements\n\n* `Keras >= 2.2.0` / `TensorFlow >= 1.12.0`\n* `keras_applications >= 1.0.7`\n* `scikit-image`\n\n### Installing from the source\n\n```bash\n$ pip install -U git+https://github.com/qubvel/efficientnet\n```\n\n### Installing from PyPI\n\nPyPI stable release\n\n```bash\n$ pip install -U efficientnet\n```\n\nPyPI latest release (with keras and tf.keras support)\n\n```bash\n$ pip install -U --pre efficientnet\n```\n\n## Frequently Asked Questions\n\n* **How can I convert the original TensorFlow checkpoints to Keras HDF5?**\n\nPick the target directory (like `dist`) and run the [converter script](./scripts) from the repo directory as follows:\n\n```bash\n$ ./scripts/convert_efficientnet.sh --target_dir dist\n```\n\nYou can also optionally create the virtual environment with all the dependencies installed by adding `--make_venv=true` and operate in a self-destructing temporary location instead of the target directory by setting `--tmp_working_dir=true`.\n\n## Acknowledgements\nI would like to thanks community members who actively contribute to this repository:\n\n1) Sasha Illarionov ([@sdll](https://github.com/sdll)) for preparing automated script for weights conversion\n2) Bj\u00f6rn Barz ([@Callidior](https://github.com/Callidior)) for model code adaptation for keras and tensorflow.keras frameworks \n\n\n",
"bugtrack_url": null,
"license": "Apache License 2.0",
"summary": "EfficientNet model re-implementation. Keras and TensorFlow Keras.",
"version": "1.1.1",
"split_keywords": [],
"urls": [
{
"comment_text": "",
"digests": {
"md5": "ef20159b40c5c5944c0a267d05e0ed3c",
"sha256": "6967fcdaab2074f96228f684dad0628febe622a59181b4a28aaca1251f1b8784"
},
"downloads": -1,
"filename": "efficientnet-1.1.1-py3-none-any.whl",
"has_sig": false,
"md5_digest": "ef20159b40c5c5944c0a267d05e0ed3c",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.0.0",
"size": 18420,
"upload_time": "2020-09-15T16:26:00",
"upload_time_iso_8601": "2020-09-15T16:26:00.362832Z",
"url": "https://files.pythonhosted.org/packages/53/97/84f88e581d6ac86dcf1ab347c497c4c568c38784e3a2bd659b96912ab793/efficientnet-1.1.1-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"md5": "dbab59272ed84a46ab608e15bb897a97",
"sha256": "a92e7715453f6043942d9c8d995464e8d319494f08d1d458600abbd5c43544bb"
},
"downloads": -1,
"filename": "efficientnet-1.1.1.tar.gz",
"has_sig": false,
"md5_digest": "dbab59272ed84a46ab608e15bb897a97",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.0.0",
"size": 15984,
"upload_time": "2020-09-15T16:26:01",
"upload_time_iso_8601": "2020-09-15T16:26:01.845208Z",
"url": "https://files.pythonhosted.org/packages/bc/0f/811c73e9e579361b202b1e8205fff114ee7f9a738489247207c9141266f3/efficientnet-1.1.1.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2020-09-15 16:26:01",
"github": true,
"gitlab": false,
"bitbucket": false,
"github_user": "qubvel",
"github_project": "efficientnet",
"travis_ci": true,
"coveralls": false,
"github_actions": false,
"requirements": [
{
"name": "keras_applications",
"specs": [
[
"<=",
"1.0.8"
],
[
">=",
"1.0.7"
]
]
},
{
"name": "scikit-image",
"specs": []
}
],
"lcname": "efficientnet"
}
JSON
