# Successes and Limitations of Object-centric Models at Compositional Generalisation

arXiv: https://arxiv.org/abs/2412.18743 · HTML: https://mlsanity.com/paper/2412.18743/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 10 of its 24 references that are in the map.
- First posted on arXiv 25 Dec 2024.

## Builds on (most cited first, 10 of 10)

- [Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges](https://mlsanity.com/paper/2104.13478/) (2021)
- [Object-Centric Learning with Slot Attention](https://mlsanity.com/paper/2006.15055/) (2020)
- [Wasserstein Auto-Encoders](https://mlsanity.com/paper/1711.01558/) (2017)
- [Illiterate DALL-E Learns to Compose](https://mlsanity.com/paper/2110.11405/) (2021)
- [Compositional Generalization from First Principles](https://mlsanity.com/paper/2307.05596/) (2023)
- [Visual Representation Learning Does Not Generalize Strongly Within the Same Domain](https://mlsanity.com/paper/2107.08221/) (2021)
- [Unsupervised Model Selection for Variational Disentangled Representation Learning](https://mlsanity.com/paper/1905.12614/) (2019)
- [Neural Systematic Binder](https://mlsanity.com/paper/2211.01177/) (2022)
- [Unsupervised Object Learning via Common Fate](https://mlsanity.com/paper/2110.06562/) (2021)
- [Learning and generalization of compositional representations of visual scenes](https://mlsanity.com/paper/2303.13691/) (2023)

## Built on by (most cited first, 2 of 2)

- [Oh-A-DINO: Understanding and Enhancing Attribute-Level Information in Self-Supervised Object-Centric Representations](https://mlsanity.com/paper/2503.09867/) (2025)
- [Are Object-Centric Representations Better At Compositional Generalization?](https://mlsanity.com/paper/2602.16689/) (2026)

## Abstract (arXiv)

In recent years, it has been shown empirically that standard disentangled latent variable models do not support robust compositional learning in the visual domain. Indeed, in spite of being designed with the goal of factorising datasets into their constituent factors of variations, disentangled models show extremely limited compositional generalisation capabilities. On the other hand, object-centric architectures have shown promising compositional skills, albeit these have 1) not been extensively tested and 2) experiments have been limited to scene composition -- where models must generalise to novel combinations of objects in a visual scene instead of novel combinations of object properties. In this work, we show that these compositional generalisation skills extend to this later setting. Furthermore, we present evidence pointing to the source of these skills and how they can be improved through careful training. Finally, we point to one important limitation that still exists which suggests new directions of research.