iSAID_split

Split iSAID dataset and its coco-format json annotation files.

Environment and dependencies installation

Create the conda environment

> conda create -n py_isaid python=3.8
> pip install pycocotools opencv-python natsort h5py scipy

Dataset preparation

Download the iSAID dataset from iSAID.
Unzip the dataset to the data folder.
Make sure that the final dataset must have this structure:

iSAID
├── test
│   └── images
│       ├── P0006.png
│       └── ...
│       └── P0009.png
├── train
│   └── images
│       ├── P0002_instance_color_RGB.png
│       ├── P0002_instance_id_RGB.png
│       ├── P0002.png
│       ├── ...
│       ├── P0010_instance_color_RGB.png
│       ├── P0010_instance_id_RGB.png
│       └── P0010.png
└── val
    └── images
        ├── P0003_instance_color_RGB.png
        ├── P0003_instance_id_RGB.png
        ├── P0003.png
        ├── ...
        ├── P0004_instance_color_RGB.png
        ├── P0004_instance_id_RGB.png
        └── P0004.png

Run the following command to split the dataset:

> python split.py --set train,val
> python split.py --set test

Note that "train","val" cannot be split simultaneously with "test". If --set includes "test", the "train" and "val" will be split just image without annotation.

Run the following command to split the dataset json file:

python preprocess.py --set train,val

Make sure that the final dataset after preprocesing must have this structure:

iSAID_patches
├── test
│   └── images
│       ├── P0006_0_0_800_800.png
│       └── ...
│       └── P0009_0_0_800_800.png
├── train
│   └── instance_only_filtered_train.json
│   └── images
│       ├── P0002_0_0_800_800_instance_color_RGB.png
│       ├── P0002_0_0_800_800_instance_id_RGB.png
│       ├── P0002_0_800_800.png
│       ├── ...
│       ├── P0010_0_0_800_800_instance_color_RGB.png
│       ├── P0010_0_0_800_800_instance_id_RGB.png
│       └── P0010_0_800_800.png
└── val
    └── instance_only_filtered_val.json
    └── images
        ├── P0003_0_0_800_800_instance_color_RGB.png
        ├── P0003_0_0_800_800_instance_id_RGB.png
        ├── P0003_0_0_800_800.png
        ├── ...
        ├── P0004_0_0_800_800_instance_color_RGB.png
        ├── P0004_0_0_800_800_instance_id_RGB.png
        └── P0004_0_0_800_800.png

Here is a sample dataset to test this project.

If you want to change the folder for reading and saving image, please modify the parameters --src, --tar in split.py and --outdir, --datadir in preprocess.py.

If you want to change the size and the overlap area of the split please modify to the parameters --patch_width, --patch_height, --overlap_area in split.py.

If you want to delete the images with empty annotation, you can use clear_isaid.py to clear the folder by python clear_isaid.py --clear-folder <clear_folder_path>.

Acknowledgement

This script is developed based on iSAID_Devkit.

songbingze/iSAID_split

iSAID_split

Acknowledgement