Image Segmentation-based Unsupervised Multiple Objects Discovery

TL;DR

This paper introduces a fully unsupervised, two-stage bottom-up approach for discovering multiple objects in images, leveraging self-supervised features and dataset-wide semantic models, achieving state-of-the-art results.

Contribution

It presents a novel unsupervised framework that combines intra-image similarity and dataset-level semantic modeling for object discovery without annotations.

Findings

Outperforms existing methods in precision-recall trade-off.

Achieves state-of-the-art results in unsupervised object detection.

Effective in unsupervised image segmentation.

Abstract

Unsupervised object discovery aims to localize objects in images, while removing the dependence on annotations required by most deep learning-based methods. To address this problem, we propose a fully unsupervised, bottom-up approach, for multiple objects discovery. The proposed approach is a two-stage framework. First, instances of object parts are segmented by using the intra-image similarity between self-supervised local features. The second step merges and filters the object parts to form complete object instances. The latter is performed by two CNN models that capture semantic information on objects from the entire dataset. We demonstrate that the pseudo-labels generated by our method provide a better precision-recall trade-off than existing single and multiple objects discovery methods. In particular, we provide state-of-the-art results for both unsupervised class-agnostic object detection and unsupervised image segmentation.