Zero-Shot Object-Centric Representation Learning

TL;DR

This paper investigates zero-shot generalization in object-centric representation learning, introduces a benchmark with diverse datasets, and proposes a fine-tuning strategy that achieves state-of-the-art results in unsupervised object discovery.

Contribution

It introduces a new benchmark for zero-shot object-centric learning and a fine-tuning method that enhances transferability and performance on unseen datasets.

Findings

Training on diverse real-world images improves zero-shot transfer.

The proposed fine-tuning strategy achieves state-of-the-art unsupervised object discovery.

Strong zero-shot transfer performance on unseen datasets.

Abstract

The goal of object-centric representation learning is to decompose visual scenes into a structured representation that isolates the entities. Recent successes have shown that object-centric representation learning can be scaled to real-world scenes by utilizing pre-trained self-supervised features. However, so far, object-centric methods have mostly been applied in-distribution, with models trained and evaluated on the same dataset. This is in contrast to the wider trend in machine learning towards general-purpose models directly applicable to unseen data and tasks. Thus, in this work, we study current object-centric methods through the lens of zero-shot generalization by introducing a benchmark comprising eight different synthetic and real-world datasets. We analyze the factors influencing zero-shot performance and find that training on diverse real-world images improves transferability to unseen scenarios. Furthermore, inspired by the success of task-specific fine-tuning in foundation models, we introduce a novel fine-tuning strategy to adapt pre-trained vision encoders for the task of object discovery. We find that the proposed approach results in state-of-the-art performance for unsupervised object discovery, exhibiting strong zero-shot transfer to unseen datasets.