A Critical Robustness Evaluation for Referring Expression Comprehension Methods
zhipeng zhang (Northwestern Polytechnical University), Zhimin Wei (Northwestern Polytechnical University), Peng Wang (Northwestern Polytechnical University)*
The 34th British Machine Vision Conference
Abstract
Referring Expression Comprehension (REC) is a crucial task in visual reasoning that requires models to accurately identify target objects indicated by natural language expressions. Researchers have focused on the performance of models on the COCO dataset (RefCOCO/RefCOCO+/RefCOCOg) and employed various training strategies to improve performance scores. However, there is a lack of robustness evaluation and analysis among these works due to the absence of an evaluation metric and dataset benchmarks for comparison. In this work, we propose a novel dataset and benchmark for the word-level adversarial robustness of Referring Expression Comprehension task. We also evaluate the robustness experiments on several previous strong methods.Video
Citation
@inproceedings{zhang_2023_BMVC,
author = {zhipeng zhang and Zhimin Wei and Peng Wang},
title = {A Critical Robustness Evaluation for Referring Expression Comprehension Methods},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year = {2023},
url = {https://papers.bmvc2023.org/0045.pdf}
}
Copyright © 2023
The British Machine Vision Conference is organised by