Roof World (+ GUI)

Inferring 3D structure from 2D graphs.

Quickstart

# python >= 3.10
pip install -r requirements.txt

Run the GUI and interact with the model

python gui.py

Basic GUI Usage:

Left click to add a point
Right click on a point to start drawing a line, Right click on another point to connect them
Press p on your keyboard to run prediction and produce a 3D plot showing the probability of each point being a "high point"

Other useful features:

Left click on a point to toggle black/orange color (orange represents a high point, black represents "all other points") for creating training data
CTRL-z: Undo
ESCAPE: Quit "line drawing" mode
c: (C)lear all points and lines
s: (S)ave the current state to the saved/ directory
b: (B)ackground toggle (turn the graph paper on or off, useful for taking screenshots)
d: (D)ump internal state to the console, showing the predicate logic representation

Estimate cross-validation performance

The saved/ directory contains 20 building layouts. Running estimate_performance.py estimates 5-fold cross-validation for predicting whether a point is a high point.

python estimate_performance.py

Other useful features

Finding points and lines in an image with Hough transform