https://github.com/devbruce/cutmiximagedatagenerator_for_keras

Keras implementation of CutMix regularizer
https://github.com/devbruce/cutmiximagedatagenerator_for_keras

augmentation cutmix generator keras regularizer

Last synced: about 2 hours ago
JSON representation

Keras implementation of CutMix regularizer

• Host: GitHub
• URL: https://github.com/devbruce/cutmiximagedatagenerator_for_keras
• Owner: devbruce
• License: mit
• Created: 2019-08-06T09:47:18.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2020-02-11T01:22:44.000Z (almost 5 years ago)
• Last Synced: 2024-11-09T05:46:47.968Z (10 days ago)
• Topics: augmentation, cutmix, generator, keras, regularizer
• Language: Python
• Homepage:
• Size: 8.79 KB
• Stars: 29
• Watchers: 3
• Forks: 12
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# CutMixImageDataGenerator (Keras)

![GitHub release (latest by date)](https://img.shields.io/github/v/release/DevBruce/CutMixImageDataGenerator_For_Keras)

> Paper: [CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features](https://arxiv.org/abs/1905.04899)

## Citation

```

@misc{yun2019cutmix,

    title={CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features},

    author={Sangdoo Yun and Dongyoon Han and Seong Joon Oh and Sanghyuk Chun and Junsuk Choe and Youngjoon Yoo},

    year={2019},

    eprint={1905.04899},

    archivePrefix={arXiv},

    primaryClass={cs.CV}

}

```




## Install

```bash

$ pip install cutmix-keras

```




## Using Example

```python

# (some codes) ...

from cutmix_keras import CutMixImageDataGenerator  # Import CutMix

train_datagen = ImageDataGenerator(

    rescale=1./255,

)

train_generator1 = train_datagen.flow_from_dataframe(

    dataframe=X_train,

    directory=IMG_DIR,

    target_size=(IMG_SIZE, IMG_SIZE),

    x_col='X_Column',

    y_col='Y_Column',

    color_mode='rgb',

    class_mode='categorical',

    batch_size=BATCH_SIZE,

    shuffle=True,  # Required

)

train_generator2 = train_datagen.flow_from_dataframe(

    dataframe=X_train,

    directory=IMG_DIR,

    target_size=(IMG_SIZE, IMG_SIZE),

    x_col='X_Column',

    y_col='Y_Column',

    color_mode='rgb',

    class_mode='categorical',

    batch_size=BATCH_SIZE,

    shuffle=True,  # Required

)

# !! Define CutMixImageDataGenerator !!

train_generator = CutMixImageDataGenerator(

    generator1=train_generator1,

    generator2=train_generator2,

    img_size=IMG_SIZE,

    batch_size=BATCH_SIZE,

)

# (some codes) ...

history = model.fit_generator(

        generator=train_generator,

        steps_per_epoch=train_generator.get_steps_per_epoch(),

        # (some parameters) ...

        )

```




### Example of Kaggle Kernel

Link: [[KaKR\_2019\_3rd] CutMix, Ensemble (Keras)](https://www.kaggle.com/devbruce/kakr-2019-3rd-cutmix-ensemble-keras#Generator)




- `generator1`, `generator2` need same generator applied `flow` method




- `generator1`, `generator2` need `shuffle=True`  

If `shuffle=False`, This generator cutmix with same images.  

So there would no augmentation




- Why are there two same generators? (`generator1`, `generator2`)  

\-\-\> To Solve Reference Problem