Skip to content

Latest commit

 

History

History
33 lines (28 loc) · 1.21 KB

README.md

File metadata and controls

33 lines (28 loc) · 1.21 KB
Raw

smart-labeller

Labelling tool for creating custom datasets for object detection and semantic / instance segmentation. Intended to make labelling segmentation mask easier for abstract objects (e.g. cracks or erosion on surfaces) using algorithms such as GrabCut and image thresholding.

demo
Example usage: labelling defects on a wind turbine's blade surface using this repository

Installation

git clone https://github.com/nearthlab/smart-labeller
cd smart-labeller
pip install -r requirements.txt

Usage

First, prepare images and class_labels.json as in the datasets/sample.

  1. Create labels for new images or edit existing labels
python label.py [(optional) /path/to/dataset]
# press F1 to see instructions
  1. Augment existing labels
python augment.py [(optional) /path/to/dataset]
# set options as you wish and press augment button
  1. Export your labels into one of PASCAL-VOC / COCO / CityScapes (a.k.a. KITTI) dataset.
python export.py [(optional) /path/to/dataset]
# set options as you wish and press export button