CLOT: Closed Loop Optimal Transport for Unsupervised Action Segmentation

Elena Bueno-Benito, Mariella Dimiccoli · 2025

Read CLOT: Closed Loop Optimal Transport for Unsupervised Action Segmentation by Elena Bueno-Benito, Mariella Dimiccoli (2025) 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-2507-03539.txt.

Description

Elena Bueno-Benito and Mariella Dimiccoli cut videos into actions with no labels at all in this 2025 paper. Their method, CLOT, grows out of an optimal transport approach that learns action types and groups frames at once. That earlier method lacked feedback at the level of whole segments. CLOT closes the loop with three linked transport problems.

Frame and segment embeddings learn side by side. Cross-attention between them then refines both the features and the draft labels. The cycle repeats, each pass sharpening the last. On four benchmark datasets the closed loop beat its open-loop parent. The authors credit the gains to segment-level supervision arising from the loop itself.

Catalog record

CREATORElena Bueno-Benito, Mariella Dimiccoli
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-03539

Browse the full library · About the Tower of Babel