Towards Robust Evaluation of Visual Activity Recognition: Resolving Verb Ambiguity with Sense Clustering

Louie Hong Yao [email protected], Nicholas Jarvis [email protected], Tianyu Jiang [email protected] · 2025

Read Towards Robust Evaluation of Visual Activity Recognition: Resolving Verb Ambiguity with Sense Clustering by Louie Hong Yao [email protected], Nicholas Jarvis [email protected], Tianyu Jiang [email protected] (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-2508-04945.txt.

Description

Louie Hong Yao, Nicholas Jarvis, and Tianyu Jiang proposed in 2025 a fairer way to grade systems that name actions in images. Verbs are slippery: two words can name the same act, and one scene can honestly take several verbs. Grading against a single gold verb thus underrates good systems. Their fix groups verbs into sense clusters, so credit goes to any answer in the right cluster.

On a standard dataset, each image mapped to about four clusters, each a distinct reading of the scene. Models scored under the cluster method lined up better with human judgments than under exact matching. The authors argue cluster grading gives a fuller picture of what action recognition systems can do. The catalog record garbles the creator field with email addresses; the byline in the text is clean.

Catalog record

CREATORLouie Hong Yao [email protected], Nicholas Jarvis [email protected], Tianyu Jiang [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-04945

Browse the full library · About the Tower of Babel