spytensor/prepare_detection_dataset

What it solves

This project provides a set of scripts to handle the tedious task of converting object detection datasets between different common annotation formats. It simplifies the process of preparing data for various machine learning frameworks that require specific input formats.

How it works

The toolset consists of several Python scripts that transform data between the following formats:

• CSV to COCO: Converts a custom CSV format (containing image paths, bounding box coordinates, and labels) into the the COCO dataset structure.
• CSV to VOC: Converts CSV annotations into the Pascal VOC XML format.
• Labelme to COCO: Transforms JSON annotations created with the Labelme tool into COCO format.
• Labelme to VOC: Transforms Labelme JSON annotations into Pascal VOC XML format.
• CSV to Labelme: Converts CSV data back into Labelme-compatible JSON files.

Who it’s for

Developers and researchers working on object detection projects who need to move their annotations between different labeling tools or training frameworks.

Highlights

• Cross-format support: Handles conversions between CSV, COCO, and Pascal VOC.
• Labelme integration: Direct support for converting to and from the Labelme annotation tool.
• Automated organization: Automatically creates the necessary directory structures and copies images to the target format's required layout.