O-SE-R

This repository contains the implementation details for the OCR tool built to convert UML sketches to a code skeleton using Image Processing.

How to use O-SE-R

Prerequisites

• opencv: 4.6.0.66
• numpy: 1.23.2
• Download the custom YOLO weight file here: https://drive.google.com/drive/folders/1Bp7WFeSSX3HlyC1KIU4BPRpBHqo5AbLq?usp=sharing

Run O-SE-R

• Detect the UML classes and inheritance symbols: yolo_img.py
• Generate code skeleton: gen_code_skel.py

Test Set Results:

  Total Number of Classes: 274
  Number of Classes Identified: 233
  % of Classes Identified: 85.03%

  Total Number of Inheritance Symbols: 50
  Number of Inheritance Symbols Identified: 44
  % of Classes Identified: 88.00%

  Total Number of Associations: 144
  Number of Associations Identified: 64
  % of Associations Identified: 44.44%

  Total Number of Generalizations: 50
  Number of Generalizations Identified: 18
  % of Generalizations Identified: 36%

Dataset

An open-source dataset was annotated to the train the YOLO model on UML classes and inheritance symbols.

Original dataset link: https://www.kaggle.com/datasets/leticiapiucco/handwritten-uml-class-diagrams