Datasets tools

This is a bunch of simple python scripts to convert between different datasets` annotation formats.

Usage

Caltech to pascalVoc

python matlab_to_pacal_voc.py <path to .xml file> <path to output directory>

This script converts Caltech dataset`s annotations from matlab array to pascalVoc.
It will save .xml file for every image to annotations directory in <path to output directory>.

Yolo to pascalVov

python yolo_to_voc.py <path to directory with annotations in yolo format> <path to output direcotry> <path to directory with images>

Currently this script supports only one class and you can change it on 49 line.

PascalVoc to COCO

python pascalVoc2coco.py <path to direcoty with pascalVoc annotations> <path to output json file>

This script converts annotations from pascalVoc format to coco.
You can change info about your dataset in script.

SoF to json

python sof_to_json.py <path to .mat input file> <path to json output file>

This script converts SoF annotations from matlab array to json in this format:

[
    {
        "id": "AbdA",
        "sequence": "00001",
        "gender": "m",
        "age": 31,
        "lightning": "i",
        "view": "fr",
        "cropped": "nc",
        "emotion": "no",
        "year": 2016,
        "part": "2",
        "glasses": 1,
        "headscarf": false,
        "illumination": true,
        "filename": "AbdA_00001_m_31_*",
        "landmarks": [17 elemts: {"x": x, "y": y}],
        "estimated_landmarks": [17 bools],
        "face_ROI": {"x": x, "y": y, "width": width, "height": height},
        "glasses_ROI": {"x": x, "y": y, "width": width, "height": height}
      }
    }
]

Rename pascalVoc images

python rename_pscalVoc_images.py <directory with pascalVoc annotations>

This script will rename images with annotations in pascalVoc format with numbers.
Maybe it is useless but it makes dataset look prettier.