Self-Supervised Transformers for Unsupervised Object Discovery using Normalized Cut

TL;DR

This paper introduces a graph-based method leveraging self-supervised transformer features and normalized graph-cut for unsupervised object discovery, significantly improving performance over previous methods.

Contribution

The paper presents a novel graph-based approach using spectral clustering on transformer features for unsupervised object discovery, enhancing accuracy and extending to saliency detection and weakly supervised detection.

Findings

Improves unsupervised object discovery performance by up to 8.1% on VOC and COCO datasets.

Enhances unsupervised saliency detection IoU scores on ECSSD, DUTS, DUT-OMRON datasets.

Achieves competitive results in weakly supervised object detection on CUB and ImageNet.

Abstract

Transformers trained with self-supervised learning using self-distillation loss (DINO) have been shown to produce attention maps that highlight salient foreground objects. In this paper, we demonstrate a graph-based approach that uses the self-supervised transformer features to discover an object from an image. Visual tokens are viewed as nodes in a weighted graph with edges representing a connectivity score based on the similarity of tokens. Foreground objects can then be segmented using a normalized graph-cut to group self-similar regions. We solve the graph-cut problem using spectral clustering with generalized eigen-decomposition and show that the second smallest eigenvector provides a cutting solution since its absolute value indicates the likelihood that a token belongs to a foreground object. Despite its simplicity, this approach significantly boosts the performance of unsupervised object discovery: we improve over the recent state of the art LOST by a margin of 6.9%, 8.1%, and 8.1% respectively on the VOC07, VOC12, and COCO20K. The performance can be further improved by adding a second stage class-agnostic detector (CAD). Our proposed method can be easily extended to unsupervised saliency detection and weakly supervised object detection. For unsupervised saliency detection, we improve IoU for 4.9%, 5.2%, 12.9% on ECSSD, DUTS, DUT-OMRON respectively compared to previous state of the art. For weakly supervised object detection, we achieve competitive performance on CUB and ImageNet.