FastGDP

Introduction

This repository contains the code and data for the paper "An Efficient Approach to Automated Geometry Diagram Parsing" published in the Journal of Emerging Investigators.

FastGDP stands for Fast Geometry Diagram Parser. Given a geometry diagram, FastGDP can recognize lines, circles and points. Additionally, FastGDP can determine which lines or circles a point lies on and of a point is the center of a circle. FastGDP can also detect text labels and associate them with points, but this functionality does not currently work very well.

FastGDP is written to be modular which means that it is easy to make changes to the various components in order to improve them or customize them for a particular application.

Quickstart

• Clone or download this repository

• Open a command prompt window and install OpenCV as follows:

  pip install opencv-python

• Import OpenCV as follows:

  import cv2

• Import parse_diagram and display_interpretation from diagram_parser.diagram_graph_builder as follows:

  from diagram_parser.diagram_graph_builder import parse_diagram, display_interpretation

• Load your diagram image using OpenCV (the diagram image should be in the BGR format) as follows:

  img = cv2.imread('<image_path>')

• Call parse_diagram, passing the diagram image as follows:

  interpretation, lines, circles = parse_diagram(img)

  The parse_diagram method returns a tuple (interpretation, lines, circles). lines and circles are both dictionaries. interpretation is an instance of the Interpretation class.

• Optionally call display_interpretation to visualize the detection result, passing the original diagram image, interpretation, lines, and circles:

  display_interpretation(img, interpretation, lines, circles)

You can find more information on how to use the results of the parse_diagram method below.

Note that FastGDP does not give good results if the diagram image is either very small or very large. The best results are obtained when the width and height of the image are between 150 and 350 pixels.

How to use the results of parse_diagram

parse_diagram returns a tuple (interpretation, lines, circles).

Interpretation

An Interpretation object contains the instance variables points, lines, and circles. points is a list of instances of the Point class. lines and circles are both dictionaries. Each key in lines is the ID of a line (used to refer to that line in point properties) and the corresponding value is that line in Hesse normal form. Each key in circles is the ID of a circle and the corresponding value of the circle specified in the (x, y, r) format where (x, y) are the coordinates of the center and r is the radius.

Point

Iterating over interpretation in a for loop will yield the points in the interpretation one at a time. Each Point object has instance variables label, which is the label of the point, coords, which are the coordinates of the point in the image, and properties, which is a list of the properties of the point. If a label is detected for a point, the label is a single uppercase letter. Otherwise, the label is of the form p<x> where x is an integer.

Point properties

Each point property is a tuple where the first element is either the string "lieson" or the string "centerof" and the second element is the ID of either a line or a circle. If the second element refers to a line, the ID is of the form l<x>, where x is an integer. The lines dictionary will contain l as a key and the corresponding value will be the line in Hesse normal form. If the second element of the property tuple refers to a circle, the ID is of the form c, where c is an integer. The circles dictionary will contain x as a key, and the corresponding value will be the circle in the (x, y, r) format.

Datasets

Three datasets are provided in this repository. The first is the training data used by geosolver. The second is a dataset containing significantly more complex data. The third is the geosolver test dataset. The images and annotations for each dataset can be found in the experiments/data directory.