Learning Causality from Video

Introduction

This code implements the theory to learn the Causal And-Or Graph from the paper "Learning Perceptual Causality from Video" by Amy Fire and Song-Chun Zhu, 2016. The associated paper and data can be found at http://amyfire.com/projects/learningcausality/.

This code uses a minimax entropy pursuit alongside heuristics to attribute strong correlations as causal.

Required Data

Source data consists of detections made in 4 scenes--3 different kinds of doorways (data/Exp1_output_data_key.txt, data/Exp1_output_data3.txt, data/Exp1_output_data2.txt) and 1 office scene (data/Exp2_output_data.txt). Detections were performed by Mingtao Pei.

Workflow

Output for the paper is produced by running the script LEARNING.m. It requires the detections above in data/.