ChartLens: Fine-grained Visual Attribution in Charts

Manan Suri, Puneet Mathur, Nedim Lipka, Franck Dernoncourt, Ryan A. Rossi, Dinesh Manocha · 2025

Read ChartLens: Fine-grained Visual Attribution in Charts by Manan Suri, Puneet Mathur, Nedim Lipka, Franck Dernoncourt, Ryan A. Rossi, Dinesh Manocha (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-2505-19360.txt.

Description

Manan Suri and five co-authors present ChartLens, a tool that ties chart answers back to the chart itself. Their twenty twenty-five paper deals with multimodal models that read charts but often hallucinate, stating facts the chart does not show. The fix they propose checks each claim after the fact.

ChartLens finds the parts of a chart, bars, labels, and axes, by segmentation. It then asks the model, with marks set on those parts, which parts back a given answer. The team also built an eval bench with fake and real charts from finance, policy, and economics. ChartLens raised the quality of these fine-grained attributions by wide margins.

Catalog record

CREATORManan Suri, Puneet Mathur, Nedim Lipka, Franck Dernoncourt, Ryan A. Rossi, Dinesh Manocha
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-19360

Browse the full library · About the Tower of Babel