Spatially Grounded Explanations in Vision Language Models for Document Visual Question Answering

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lagos, Maximiliano Hormazábal, Cerezo-Costas, Héctor, Karatzas, Dimosthenis
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912487787462656
author Lagos, Maximiliano Hormazábal
Cerezo-Costas, Héctor
Karatzas, Dimosthenis
author_facet Lagos, Maximiliano Hormazábal
Cerezo-Costas, Héctor
Karatzas, Dimosthenis
contents We introduce EaGERS, a fully training-free and model-agnostic pipeline that (1) generates natural language rationales via a vision language model, (2) grounds these rationales to spatial sub-regions by computing multimodal embedding similarities over a configurable grid with majority voting, and (3) restricts the generation of responses only from the relevant regions selected in the masked image. Experiments on the DocVQA dataset demonstrate that our best configuration not only outperforms the base model on exact match accuracy and Average Normalized Levenshtein Similarity metrics but also enhances transparency and reproducibility in DocVQA without additional model fine-tuning.
format Preprint
id arxiv_https___arxiv_org_abs_2507_12490
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Spatially Grounded Explanations in Vision Language Models for Document Visual Question Answering
Lagos, Maximiliano Hormazábal
Cerezo-Costas, Héctor
Karatzas, Dimosthenis
Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
Machine Learning
We introduce EaGERS, a fully training-free and model-agnostic pipeline that (1) generates natural language rationales via a vision language model, (2) grounds these rationales to spatial sub-regions by computing multimodal embedding similarities over a configurable grid with majority voting, and (3) restricts the generation of responses only from the relevant regions selected in the masked image. Experiments on the DocVQA dataset demonstrate that our best configuration not only outperforms the base model on exact match accuracy and Average Normalized Levenshtein Similarity metrics but also enhances transparency and reproducibility in DocVQA without additional model fine-tuning.
title Spatially Grounded Explanations in Vision Language Models for Document Visual Question Answering
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
Machine Learning
url https://arxiv.org/abs/2507.12490