data_parsers

Collection of data parsers in different formats for training neural network models on radioastronomical datasets Parses a single split of the dataset, so the split has to be preventively done when running this script

COCO Parser

Converts FITS mask data in COCO format

YOLO Parser

Converts FITS mask data in YOLO format

Args

• -p Type of parser (default: coco)
• -m, Path of file that lists all mask file paths (trainset.dat)
• -d Destination directory for converted data (default: coco)
• -c Contrast value for conversion from FITS to PNG (default: 0.15)

Usage

Extract the whole content of the MLDataset_cleaned.tar.gz archive, perform the split using the data_splitter script, then put the folders in the same parent directory (next section provides a visual hint) and launch sh process_splits.sh to run the script for all splits. The result will be stored in a directory named coco (or in another one if specified)

Directory Structure

parent_folder
└───data_parsers
│   │───main.py
│   │───...
│   │───README.md (**YOU ARE HERE**)    
│   │
└───data_dir (e.g. MLDataset_cleaned)
    │
    └───train
    │   │───imgs
    │   │───annotations
    │   │───masks
    │   │───imgs_png
    │   │───...
    └───val
    │   │───imgs
    │   │───annotations
    │   │───masks
    │   │───imgs_png
    │   │───...
    └───test
        │───imgs
        │───annotations
        │───masks
        │───imgs_png
        │───...