Saved in:
Bibliographic Details
Main Authors: Hashem, Tahsina, Wang, Weiqing, Wijaya, Derry Tanti, Ali, Mohammed Eunus, Li, Yuan-Fang
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2409.03961
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917770204020736
author Hashem, Tahsina
Wang, Weiqing
Wijaya, Derry Tanti
Ali, Mohammed Eunus
Li, Yuan-Fang
author_facet Hashem, Tahsina
Wang, Weiqing
Wijaya, Derry Tanti
Ali, Mohammed Eunus
Li, Yuan-Fang
contents While large multimodal models (LMMs) have obtained strong performance on many multimodal tasks, they may still hallucinate while generating text. Their performance on detecting salient features from visual data is also unclear. In this paper, we develop a framework to generate faithful and salient text from mixed-modal data, which includes images and structured data ( represented in knowledge graphs or tables). Specifically, we train a small vision critic model to identify hallucinated and non-salient features from the image modality. The critic model also generates a list of salient image features. This information is used in the post editing step to improve the generation quality. Experiments on two datasets show that our framework improves LMMs' generation quality on both faithfulness and saliency, outperforming recent techniques aimed at reducing hallucination.
format Preprint
id arxiv_https___arxiv_org_abs_2409_03961
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Generating Faithful and Salient Text from Multimodal Data
Hashem, Tahsina
Wang, Weiqing
Wijaya, Derry Tanti
Ali, Mohammed Eunus
Li, Yuan-Fang
Computer Vision and Pattern Recognition
While large multimodal models (LMMs) have obtained strong performance on many multimodal tasks, they may still hallucinate while generating text. Their performance on detecting salient features from visual data is also unclear. In this paper, we develop a framework to generate faithful and salient text from mixed-modal data, which includes images and structured data ( represented in knowledge graphs or tables). Specifically, we train a small vision critic model to identify hallucinated and non-salient features from the image modality. The critic model also generates a list of salient image features. This information is used in the post editing step to improve the generation quality. Experiments on two datasets show that our framework improves LMMs' generation quality on both faithfulness and saliency, outperforming recent techniques aimed at reducing hallucination.
title Generating Faithful and Salient Text from Multimodal Data
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2409.03961