README.md

Blues

This python library is useful for creating deep learning models, supporting Classification, Object Detection, and Semantic Segmentation, which can be evaluated or inferred using cross-validation.

Table of Contents

1. Features
2. Installation
3. Examples

Features

• Optimized for high performance
• Easy to apply cross validation
• Easy to conduct augmentations experiment
• Easy to train the following sota models
  • EfficinetNet
  • MobileNet v2 or v3
  • ResNext
  • WideResNet
  • EfficientDet
  • Shelfnet

Installation

UNDER CONSTRUCTION...

Examples

A standard deep learning situation.

# Define Data Augmentations
seq = iaa.Sequential([
    iaa.Crop(),
    iaa.Fliplr(0.5),
    iaa.GaussianBlur(sigma=(0, 3.0)),
    iaa.Cutout(),
    iaa.Multiply()
])
augmentor = blues.augmentors.ClassificationDataAugmentor(seq)

# Define Models
learning_dir = {
    'fold1': blues.models.classifications.ResNext(num_classes),
    'fold2': blues.models.classifications.WideResNet(num_classes),
    'fold3': blues.models.classifications.MobileNetV2(num_classes),
}
training_table = blues.tables.TrainingTable(learning_dir)

# Define a Dataset
dataset = blues.datasets.ClassificationDataset(
    dummy_inputs,
    dummy_teachers,
    batch_size,
    blues.resizer.ClassificationResizer((width, height)),
    transformers=transformers,
    augmentor=augmentor
)

# RUN!!!
trainer = blues.trainers.XTrainer(
    training_table,
    dataset,
    epoch,
    result_path,
    blues.metrics.accuracy,
    callback_functions=callback_functions,
    evaluate=True
)
trainer.run()