Relation Network

Overview

IPython Notebook showing pytorch implementation of Google DeepMind paper on Relation Network [1] for Sort-of-Clevr dataset.

Dependencies

numpy 
torch
os
pickle
matplotlib
sklearn
cv2
argparse

Result

[1] Screenshots

Relational	Non-Relational

[2] Test Accuracy :
Relational: 92% , Non-relational: 99%

Saved model and dataset

Google Drive Link : https://drive.google.com/drive/folders/1fdCrT2Ro3dHQk-s_Ql7zNqmDuJaUtvZd?usp=sharing
[1] Unzip data.zip into a folder "data" in the root directory of project.
[2] Put "trained_model.pth" into a folder "model" in the root directory of project.

Note

Network parameters and architecture were taken from the supplementary material[2].

Reference

[1] Santoro, Adam, et al. "A simple neural network module for relational reasoning." Advances in neural information processing systems. 2017.
[2] https://papers.nips.cc/paper/7082-a-simple-neural-network-module-for-relational-reasoning-supplemental.zip