Relation Network

Sort-of-CLEVR Dataset

Relation Network is a plug-n-play module to incorporate relational reasoning abilities to deep learning models.

This repository implements Relation Network module in PyTorch for the Sort-of-CLEVR dataset

Requirements

• PyTorch
• OpenCV
• Numpy
• Pickle
• Tqdm

Usage

1. Generate the dataset by running the data_generator.py script.
2. Start the training using the train.py script. This script will automatically evaluate on the test set at the end of the training.
3. To make predictions on new data, refer [Relation Networks_Sort-of-CLEVR.ipynb](Relation Networks_Sort-of-CLEVR.ipynb).
4. Alternatively,

References

• A simple neural network module for relational reasoning - https://arxiv.org/abs/1706.01427
• https://github.com/kimhc6028/relational-networks
• https://github.com/mesnico/RelationNetworks-CLEVR