Unsupervised Salient Object Detection With Spectral Cluster Voting

Gyungin Shin, Samuel Albanie, Weidi Xie; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2022, pp. 3971-3980

In this paper, we tackle the challenging task of unsupervised salient object detection (SOD) by leveraging spectral clustering on self-supervised features. We make the following contributions: (i) We revisit spectral clustering and demonstrate its potential to group the pixels of salient objects across various self-supervised features, e.g., MoCov2, SwAV, and DINO; (ii) Given mask proposals from multiple applications of spectral clustering on image features computed from different self-supervised models, we propose a simple but effective winner-takes-all voting mechanism for selecting the salient masks, leveraging object priors based on framing and distinctiveness; (iii) Using the selected object segmentation as pseudo groundtruth masks, we train a salient object detector, termed SelfMask, which outperforms prior approaches on three unsupervised SOD benchmarks. Code is publicly available at https://github.com/NoelShin/selfmask.