SPARK : S tochastic P ropagation via A ffinity-guided R andom wal K for training-free unsupervised segmentation
Kunal Mahatha, Jose Dolz, Christian Desrosiers · 2026
Read SPARK : S tochastic P ropagation via A ffinity-guided R andom wal K for training-free unsupervised segmentation by Kunal Mahatha, Jose Dolz, Christian Desrosiers (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-00516.txt.
Description
Kunal Mahatha and colleagues rethink training-free segmentation, arguing that spectral partitioning over diffusion affinities forces a set cluster count, blurs edges, and fails on noisy affinities. They recast segmentation as a flow equilibrium problem over affinity graphs, where stochastic dynamics blend global focus with local neighbourhoods. A Markov scheme then spreads labels by random walk, cutting weak links while backing strong affinity paths. Across seven benchmarks the method reaches state-of-the-art zero-shot results, with sharper edges and more stable masks than spectral-clustering rivals.
Catalog record
| CREATOR | Kunal Mahatha, Jose Dolz, Christian Desrosiers |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36705 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.00516