CONDA: Condensed Deep Association Learning for Co-Salient Object Detection

TL;DR

This paper introduces CONDA, a novel deep association learning framework for co-salient object detection that explicitly models inter-image associations using deep networks, hyperassociations, and a correspondence-induced condensation module, achieving superior results.

Contribution

It proposes a deep association learning strategy with hyperassociations and a condensation module, improving inter-image association modeling in co-salient object detection.

Findings

Outperforms existing methods on three benchmark datasets.

Effectively reduces noise and computational load in association modeling.

Achieves state-of-the-art accuracy across various training settings.

Abstract

Inter-image association modeling is crucial for co-salient object detection. Despite satisfactory performance, previous methods still have limitations on sufficient inter-image association modeling. Because most of them focus on image feature optimization under the guidance of heuristically calculated raw inter-image associations. They directly rely on raw associations which are not reliable in complex scenarios, and their image feature optimization approach is not explicit for inter-image association modeling. To alleviate these limitations, this paper proposes a deep association learning strategy that deploys deep networks on raw associations to explicitly transform them into deep association features. Specifically, we first create hyperassociations to collect dense pixel-pair-wise raw associations and then deploys deep aggregation networks on them. We design a progressive association generation module for this purpose with additional enhancement of the hyperassociation calculation. More importantly, we propose a correspondence-induced association condensation module that introduces a pretext task, i.e. semantic correspondence estimation, to condense the hyperassociations for computational burden reduction and noise elimination. We also design an object-aware cycle consistency loss for high-quality correspondence estimations. Experimental results in three benchmark datasets demonstrate the remarkable effectiveness of our proposed method with various training settings.