Skip to content

Latest commit

 

History

History

semantic_tools

README.md

Panoptic Segmentation Scripts

Introduction

Scripts in this repository are used to convert labelme-annotated jsons into standard datasets in PASCAL VOC format or MS COCO format.

Attention: These scripts are only for the conversion of segmentation data.

About the scripts

  • COCO_Image_Viewer.ipynb：COCO-format visualization
  • labelme2coco.py：Convert a folder of jsons into coco-format segmentation dataset:
  • segmentation_labelme2voc.py：Convert a folder of jsons into voc-format segmentation dataset and visualization pictures
  • image.py: Some simple functions。

Getting Started

labelme:

# python3
conda create --name=labelme python=3.6
conda activate labelme
pip3 install labelme==3.5.0

Other tools：

pip3 install progressbar2    # 进度条
pip3 install scikit-learn  # 用于分割数据集 
pip3 install xmltodict
pip3 install lxml
pip3 install opencv-python==3.4.9.31
pip3 install PyQt5-sip
pip3 install PyQt5

Usage

convert a folder of jsons into voc-format segmentation dataset and visualization pictures：

python segmentation_labelme2voc.py seg_labels_file jsons_dir voc_dir

  • seg_labels_file：labels json file
  • jsons_dir：json dictionary
  • voc_dir：output dictionary

Example：

python segmentation_labelme2voc.py test/label_names.txt test/test_jsons test/test_voc

convert a folder of jsons into coco-format segmentation dataset:

python labelme2coco.py

Acknowledgments

labelme2Datasets

Labelme2Coco