Tensor-Train Unsupervised Image Segmentation
Abstract
We propose TT-Seg, an unsupervised image segmentation framework that employs Tensor Train (TT) decomposition and probabilistic tensor sampling to optimize Quadratic Unconstrained Binary Optimization (QUBO) problems. TT-Seg achieves segmentation performance comparable to classical solvers while offering enhanced scalability. Experimental results indicate that the TT-based approach performs effectively on small-scale problems, although for larger QUBO instances, leading solvers such as Gurobi and the D-Wave hybrid solver remain superior.
Cite
Text
Salloum et al. "Tensor-Train Unsupervised Image Segmentation." ICLR 2025 Workshops: FPI, 2025.Markdown
[Salloum et al. "Tensor-Train Unsupervised Image Segmentation." ICLR 2025 Workshops: FPI, 2025.](https://mlanthology.org/iclrw/2025/salloum2025iclrw-tensortrain/)BibTeX
@inproceedings{salloum2025iclrw-tensortrain,
title = {{Tensor-Train Unsupervised Image Segmentation}},
author = {Salloum, Hadi and Sabbagh, Kamil and Orabi, Osama and Trabelsi, Amine and Lukin, Ruslan and Kholodov, Yaroslav},
booktitle = {ICLR 2025 Workshops: FPI},
year = {2025},
url = {https://mlanthology.org/iclrw/2025/salloum2025iclrw-tensortrain/}
}