PANC: Prior-Aware Normalized Cut via Anchor-Augmented Token Graphs

TL;DR

PANC is a training-free, prior-aware spectral segmentation method that produces stable, user-steerable masks for complex scenes, outperforming existing unsupervised approaches.

Contribution

It introduces a novel anchor-augmented eigenproblem that incorporates prior tokens, enhancing segmentation stability and accuracy without training.

Findings

PANC achieves +2.3% mIoU on DUTS-TE.

PANC achieves +2.8% mIoU on DUT-OMRON.

PANC achieves +8.7% mIoU on CrackForest.

Abstract

Unsupervised segmentation from self-supervised ViT patches holds promise but lacks robustness: multi-object scenes confound saliency cues, and low-semantic images weaken patch relevance, both leading to erratic masks. To address this, we present Prior-Aware Normalized Cut (PANC), a training-free method that data-efficiently produces consistent, user-steerable segmentations. PANC extends the Normalized Cut algorithm by connecting labeled prior tokens to foreground/background anchors, forming an anchor-augmented generalized eigenproblem that steers low-frequency partitions toward the target class while preserving global spectral structure. With prior-aware eigenvector orientation and thresholding, our approach yields stable masks. Spectral diagnostics confirm that injected priors widen eigengaps and stabilize partitions, consistent with our analytical hypotheses. PANC outperforms strong unsupervised and weakly supervised baselines, achieving mIoU improvements of +2.3% on DUTS-TE, +2.8% on DUT-OMRON, and +8.7% on low-semantic CrackForest datasets.