image-augs 0.0.2

This is a bounding box level image augmentation tool.

Image Augmentations

This is a bounding box level image augmentation tool, it can perform 14 annotations. The important ones are rotation, affine, zooming in and out, noise, and blur. The augmentations were applied to a fraction of the data (40 - 50 percent of the images out of 100 can be augmented).When rotating or zooming in and out, the bounding box coordinates will also change as the image is rotated or zoomed.

Create a folder first, inside that folder keep your image annotation folder.

Create a virtual environment

pip install virtualenv
#name your environment
python3 -m venv <your env name>

#activate the environment --> for linux user
source <your env name>/bin/activate

#activate the environment  --> for windows user
<your env name>/Source/activate.ps1

Installation (for pip installation)

pip install image_augs

After installation

create a .py script inside your created folder

#import these modules in your created <scriptname>.py file
from image_augs import utils_py
from image_augs import converter_for_txtToYolo , combined
from image_augs import draw_bbox

# first step -- >   create a destination folder to save augmented results.
saved_folder_name =  utils_py.folder_creation("<give any folder name>") 
    
#2nd step --> # give your image annotation folder(images , .txt , classes.txt)
output = converter_for_txtToYolo.converter('<give your image annotation folfer path>',keep_aspect_ratio=True,resize_im=640)
    
#3rd step -->
#different augmentation and their fraction  -- > 100 images dataset ,  if True and fraction =0.3 --> it will take random 30 images from your dataset
#raw images True means, it will save raw images too.
    
dicc = combined.main(folder=saved_folder_name,

    raw_images_ok=True,
    train_test_split=0.10,
    blurs=True,blur_f=.2,
    noise=True,noise_f=.5,
    NB=True,NB_f=.5,
    hue=True,hue_f=.5,
    sat=True,sat_f=.5,
    bright=False,bright_f=.5,
    contrast=False , contrast_f=0.5,
    rotation=True,rotation_f=0.5,
    zoom=False,zoom_f=.8,
    affine=False,affine_f=0.7,
    translation=False,translation_f= 0.7,
    vertical_flip=False,vertical_f=0.2)

#results will be saved in < your given folder >

Github (if you clone the repo from github)

Use github to clone image_augmentations repo.

git clone [e-mail]:Souviksaha1998/Image_augmentations.git
cd Image_augmentations

change in hyperparameters.ini file

---

open hyperparameters.ini * give your source folder name (image annotation folder) * destination folder, where you want to save your results , just provide any folder name * resize image, default 640. Keep aspect ratio will keep one side's (Height or width) same according to your resize image * train test split of your data * augmentations, if True, it will apply those effects. * percentage of data should use for augmentation.. 0.5 --> 50% of total data , 0.1 --> 10% of total data

Now, run this file

python mainfile.py --configFile hyperparameters.ini
#results will be saved in destination folder

To view the augmented images

aug_image = draw_bbox.bbbox_viewer('<image_path_from_train/test_folder>','<ids.pickle file path from pickle_files folder>')
cv2.imshow(f'aug_image',aug_image)
cv2.waitKey(0)

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

License

MIT