object-detection-format-converter

Dataset format converter for object detection

Overview

Deep learning field for image processing has many dataset format. For object detection task, MSCOCO, YOLO, PascalVOC and so on are frequently used and sometimes we have to convert our dataset annotation files to another. To deal with this, I creat dataset format converter for object detection. Supported dataset formats are below.

• MSCOCO
• YOLO
• PascalVOC
• KITTI

Prerequisites

• Python >= 3.8

Installation

Install python>=3.8 and run commands below in your venv or something.

pip install -U pip setuptools build
python -m build
pip install dist/objdet_converter-1.2.0-py3-none-any.whl

Usage

CLI

objdet-conv convert --src-format 'SRC_FORMAT' --dst-format 'DST_FORMAT' --src-path 'PATH_TO_SRC' --dst-path 'PAST_TO_OUTPUT' --class-txt-path 'IF NEEDED'

Supported format strings:

1. coco
2. yolo
3. pascalvoc
4. kitti

Example

objdet-conv convert --src-format yolo --dst-format coco --src-path ./yolo_dir --dst-path ./output --class-txt-path ./yolo_dir/classes.txt

Call from Function

Call 'convert_format' function with some args.

Example

from objdet_converter.convert import convert_format

convert_format(
    src_format="coco",
    dst_format="yolo",
    src_path="./coco.json",
    dst_path="./yolo_output",
    class_txt_path="",
)

Sample code is available here. Detail descriptions are here

If args of 'class_txt_path' is empty, scan all annotation files and create class list automaticaly(alphabetical order).

Detail Description

Description of Each Data Format

• MSCOCO
• YOLO
• PascalVOC
• KITTI
• Custom Dataset Implementation