Learning Perceptual Causality from Video
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
.gitignore
LEARNING.m
LICENSE
README.md
TODO.txt
addBlock.m
all_outputs_together.m
answers.m
bernoulli_KL.m
calc_KL.m
calc_info.m
chiSquare.m
createDepHGroups.m
createDepLookUp.m
create_examples_with_prev_fluent.m
dat_by_fluent_change.m
format_TEOutput.m
format_latex.m
format_table_of_infos.m
get_object_name.m
get_output_name.m
hellingerChiSquare.m
initializeDepH.m
isCorrect.m
multiple_fluents_best_time.m
multiple_fluents_best_time_increasing_randomness.m
outputIndependentBaseModel.txt
plotROC.m
plot_output.m
prepare_exp1_dat.m
pursuit.m
pushDepH.m
remove_cols.m
simVendingFull.m
simulationPursuit.m
sort_multiple_fluents.m
synthesizeDat.m
tabulate.m
updateDepH.m

README.md

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/.