https://github.com/zhoudaxia233/efficientunet
Keras Implementation of Unet with EfficientNet as encoder
https://github.com/zhoudaxia233/efficientunet
efficientnet efficientunet image-segmentation keras-tensorflow unet unet-keras
Last synced: 6 months ago
JSON representation
Keras Implementation of Unet with EfficientNet as encoder
- Host: GitHub
- URL: https://github.com/zhoudaxia233/efficientunet
- Owner: zhoudaxia233
- License: mit
- Created: 2019-06-05T17:36:24.000Z (about 6 years ago)
- Default Branch: master
- Last Pushed: 2022-04-26T12:16:50.000Z (about 3 years ago)
- Last Synced: 2024-12-02T00:49:28.749Z (7 months ago)
- Topics: efficientnet, efficientunet, image-segmentation, keras-tensorflow, unet, unet-keras
- Language: Python
- Size: 31.3 KB
- Stars: 47
- Watchers: 3
- Forks: 25
- Open Issues: 2
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# EfficientUnet
Keras Implementation of Unet with [EfficientNet](https://arxiv.org/abs/1905.11946) as encoder
- Unet with EfficientNet encoder
- EfficientNet-B0
- EfficientNet-B1
- EfficientNet-B2
- EfficientNet-B3
- EfficientNet-B4
- EfficientNet-B5
- EfficientNet-B6
- EfficientNet-B7
---
## Requirements
1. `tensorflow >= 1.13.1`
2. `Keras >= 2.2.4` (It will automatically be installed when you install `efficientunet`)
---
## Special Notice
When I built this, `tensorflow 1.13.1` and `keras 2.2.4` are the latest. There was no `TF2.0`. All the functions and the so-called "best practices" I used in this project may be obsolete. Anyway, this library still works. But please keep in mind, this is built before the advent of `TF2.0`.
---
## Installation
Install `efficientunet`:
```bash
pip install efficientunet
```
---
## Basic Usage
```bash
from efficientunet import *
model = get_efficient_unet_b5((224, 224, 3), pretrained=True, block_type='transpose', concat_input=True)
model.summary()
```
---
## Useful notes
1. This library assumes `channels_last` !
2. You cannot specify `None` for `input_shape`, since the `input_shape` is heavily used in the code for inferring
the architecture. (*The EfficientUnets are constructed dynamically*)
3. Since you cannot use `None` for `input_shape`, the image size for training process and for inference process
have to be the same.
If you do need to use a different image size for inference, a feasible solution is:
1. Save the weights of your well-trained model
2. Create a new model with the desired input shape
3. Load the weights of your well-trained model into this newly created model
4. Due to some rounding problem in the decoder path (*not a bug, this is a feature* :smirk:), the input shape should be
divisible by 32.
e.g. 224x224 is a suitable size for input images, but 225x225 is not.
---
## Potential issues
- If you are unable to load the model from the saved HDF5 file, please refer to
[this](https://github.com/keras-team/keras/issues/3867) issue.
- Especially [this](https://github.com/keras-team/keras/issues/3867#issuecomment-313336090) comment can be used as a workaround.
---
## Acknowledgment
0. Some code snippets of EfficientNet are directly borrowed from [this](https://github.com/mingxingtan/efficientnet) repo.
1. The links of pretrained weights are borrowed from [this](https://github.com/qubvel/efficientnet) repo.