whynotw / rotational-data-augmentation-yolo

rotational data augmentation for yolo

Rotational Data Augmentation for YOLO

Requirement

numpy
opencv-python

Before and after

Before: the block at left-top corner is bounded by a white box.

After: Image is rotated anticlockwise by 30 degree and the block is still bounded by the white box.

Directory structure

.
├── check_label.py
├── rotation.py
├── data_original
│   ├── images
│   │   └── test00.png
│   └── labels
│       └── test00.txt
└── data_rotational
    ├── images
    │   ├── test00_000.jpg
    │   ├── test00_030.jpg
    │   └── ...
    └── labels
        ├── test00_000.txt
        ├── test00_030.txt
        └── ...

data_original is directory with your images and labels. After running rotation.py, rotated images and labels will be stored in data_rotational.

Generate augmented data

python rotation.py DATASET_INPUT

DATASET_INPUT is data_original in this example. The default output destination directory is data_rotational.

You can use python rotation.py -h to get more information.

Visualize generated images and labels

python check_label.py DATASET_INPUT

DATASET_INPUT is data_rotational in this example.

You can use python check_label.py -h to get more information.

rotational-data-augmentation-yolo's Issues

It will generate additional error labeling

Hi,
I have used your code to generate more data, howerver, it generates an additional labeling. In my case, if my labeling amount is four of one pictrure, and it generates five labeling, and the fifith labeling value is similar with the fourth labeling value.
Can you give me some advice about this situation? Thank you