ILYA: Image Labels You Affix

A tool for labeling bounding boxes on images to be used for ML training.

Requirements

• Python 3.6

Installation

• In a terminal, run pip install git+https://github.com/robert-clayton/ILYA

How To Use

• cd into desired directory
• Have images in subfolders within Images
• Run app with ilya or python -m ilya
• Select the folder you want to label images for
• Select an image
• Click and drag a box around the desired area
• Configure label using the popup
• Repeat until finished with image
• Labels are stored in the data.csv file in the current working directory

Data Format

Taking a cue from Google's Open Images Dataset V4, boxes are saved in CSV format with the following columns:

• ImageID: String - The image this box belongs to as a relative path inside Images
• Source: String - How the box was made. Everything produced by label-maker will be xclick
• LabelName: String - What class this box is represents. Set the class list in labels.txt
• Confidence: Float - Dummy value, always 1.0
• XMin, XMax, YMin, YMax: Float - Box coordinates, normalized to the image as [0.0 ... 1.0]
• IsOccluded: Bool - Whether the object is occluding by another object
• IsTruncated: Bool - Whether the object is a part of the LabelName object
• IsGroupOf: Bool - Whether the object is a group of the LabelName object
• IsDepiction: Bool - Whether the object is a depiction of the LabelName object
• IsInside: Bool - Whether the object is the inside of the LabelName object