Unsupervised Object-Based Transition Models for 3D Partially Observable Environments

TL;DR

This paper introduces an unsupervised, object-based transition model for 3D environments that maintains object identity over time, effectively handling occlusion and partial observability without supervision.

Contribution

The proposed model uniquely combines slot-wise object decomposition, alignment, and end-to-end training to improve object persistence and identity tracking in partially observable scenes.

Findings

Outperforms state-of-the-art baseline in object tracking accuracy.

Handles occlusion and re-appearance effectively.

Operates without supervision using object-level loss.

Abstract

We present a slot-wise, object-based transition model that decomposes a scene into objects, aligns them (with respect to a slot-wise object memory) to maintain a consistent order across time, and predicts how those objects evolve over successive frames. The model is trained end-to-end without supervision using losses at the level of the object-structured representation rather than pixels. Thanks to its alignment module, the model deals properly with two issues that are not handled satisfactorily by other transition models, namely object persistence and object identity. We show that the combination of an object-level loss and correct object alignment over time enables the model to outperform a state-of-the-art baseline, and allows it to deal well with object occlusion and re-appearance in partially observable environments.