963358 / custom-dataset-tools

Tools to streamline preparing custom datasets for bounding box object detection.

Custom Dataset Tools

With the emergence of object detection in the past decade due to the unrelenting improvements in machine learning, it is more important than ever to be able to efficiently create and format datasets to easily train models. However, this can be a time consuming and confusing task because each implementation might require a different format, which is where this repository comes in.

Custom Dataset Tools contains convenient scripts to help label and format custom datasets to train machine learning object detectors. All of the scripts presume that LabelImg is used to annotate data.

Current tools:

• LabelImg fast shortcut
• Format custom datasets to:
  • COCO
  • Pascal VOC
  • YOLO
• Other miscellaneous tools

Installation

First, make sure you have Python 3 installed (only for Windows; Mac and Linux already have it preinstalled).

Open terminal and copy-paste (ctrl-shift-v for linux) the following commands:

git clone https://github.com/CraigWang1/custom-dataset-tools.git
cd custom-dataset-tools
pip3 install -r requirements.txt

To use tools, navigate to the /dataset or the /labelling directories for more instructions.

Original xml_to_json script: https://github.com/Tony607/voc2coco