Neural Causation Coefficient

An unofficial PyTorch implementation of Neural Causation Coefficient (NCC) in the paper, "Discovering Causal Signals in Images".

Requirements

torch=1.9.0
torchvision=1.10.0
scikit-learn
scipy
tqdm
numpy
pandas
matplotlib
seaborn
skimage
pycocotools

Download datasets: 1) Tubingen dataset, 2) PASCAL VOC2012 dataset, and 3) MSCOCO dataset.
NCC_Training.ipynb: Train NCC Classifier on synthetic dataset and Test it on Tubingen dataset
- Our reproduced model results/NCC_classifier_best.pt showed 78.74% accuracy on Tubingen dataset, which was 79% in the original paper.
ObjectClassifier_Training.ipynb: Train object classifier on PASCAL VOC2012 dataset
NCC_Testing.ipynb: Test NCC Classifier on MSCOCO dataset

David Lopez-Pas et al., Discovering Causal Signals in Images, ICML 2017

Name		Name	Last commit message	Last commit date
Latest commit History 29 Commits
data		data
results		results
LICENSE		LICENSE
NCC_Testing.ipynb		NCC_Testing.ipynb
NCC_Training.ipynb		NCC_Training.ipynb
ObjectClassifier_Training.ipynb		ObjectClassifier_Training.ipynb
README.md		README.md