xignos3108/data-augmentation-automated

data-augmentation-automated

Albumentation based automated data augmentation system

Contributor

ILKYU YI

How to Use

Download custom dataset(.jpg) and label file(.txt)

1. Dataset and label should exist in same folder.
2. Folder should be named as 'dataset'
3. 'dataset' folder should be located in your main directory with other .py files

Install albumentation library and dependencies

Install the latest stable version from PyPI

    pip install -U albumentations  

Install the latest version from the master branch on GitHub

    pip install -U git+https://github.com/albumentations-team/albumentations 

Force Albumentations to use dependencies

    pip install -U albumentations --no-binary qudida,albumentations

Install from conda-forge

    conda install -c conda-forge imgaug
    conda install -c conda-forge albumentations

Run main.py

    python ./main.py

Or just press run from main.py

Choose transforms and build augmentation pipline with your needs

Pipline configuration exists in configs.py

Bugs

Error occurs with 'A(0005).txt', 'E(0028).txt', 'I(0010).txt'. Remove those files before running the code.

A(0005) -> 0 0.024609 0.373611 0.049219 0.019444 -> 0.024609

(-5.000000000005e-07, 0.363889, 0.0492185, 0.38333300000000003, '0')

E(0028) -> 4 0.002734 0.859722 0.005469 0.016667 -> 0.002734

(-5.000000000000664e-07, 0.8513885, 0.005468499999999999, 0.8680555, '4')

I(0010) -> 8 0.018359 0.525694 0.036719 0.056944 -> 0.018359

(-5.000000000000664e-07, 0.8513885, 0.005468499999999999, 0.8680555, '4')

Bug fixed

1.Bbox coordinate causing error are set to be ignored in tf process (21/12/14)

2.Causing empty labels fixed (21/12/15)