CLASP: Adaptive Spectral Clustering for Unsupervised Per-Image Segmentation
It addresses the problem of unsupervised segmentation for applications like digital advertising and marketing, though it is incremental as it builds on existing methods like spectral clustering and DINO features.
The paper tackles unsupervised image segmentation by introducing CLASP, a lightweight framework that operates without labeled data or finetuning, achieving competitive mIoU and pixel accuracy on COCO Stuff and ADE20K datasets.
We introduce CLASP (Clustering via Adaptive Spectral Processing), a lightweight framework for unsupervised image segmentation that operates without any labeled data or finetuning. CLASP first extracts per patch features using a self supervised ViT encoder (DINO); then, it builds an affinity matrix and applies spectral clustering. To avoid manual tuning, we select the segment count automatically with a eigengap silhouette search, and we sharpen the boundaries with a fully connected DenseCRF. Despite its simplicity and training free nature, CLASP attains competitive mIoU and pixel accuracy on COCO Stuff and ADE20K, matching recent unsupervised baselines. The zero training design makes CLASP a strong, easily reproducible baseline for large unannotated corpora especially common in digital advertising and marketing workflows such as brand safety screening, creative asset curation, and social media content moderation