This repository contains associated code for the ICLR 2023 paper Differentiable Mathematical Programming for Object-Centric Representation Learning.

Requirements

• PyTorch
• Pytorch Lightning
• CuPy
• Scikit-sparse
• Wandb
• MediaPy
• ClevrTex dataset: https://www.robots.ox.ac.uk/~vgg/data/clevrtex/

Usage

Pretrained model checkpoint can be loaded using the Jupyter notebook for evaluation.

PYTHONPATH=. python slot_cut/train.py

Set hyperparameters in slot_cut/params.py

Acknowledgements

The repository uses code from https://github.com/untitled-ai/slot_attention and resnet code from https://github.com/kuangliu/pytorch-cifar