One does not fit all! On the Complementarity of Vision Encoders for Vision and Language Tasks

Gregor Geigle, Chen Liu, Jonas Pfeiffer, Iryna Gurevych


Abstract
Current multimodal models, aimed at solving Vision and Language (V+L) tasks, predominantly repurpose Vision Encoders (VE) as feature extractors. While many VEs—of different architectures, trained on different data and objectives—are publicly available, they are not designed for the downstream V+L tasks. Nonetheless, most current work assumes that a single pre-trained VE can serve as a general-purpose encoder. In this work, we focus on analysis and aim to understand whether the information stored within different VEs is complementary, i.e. if providing the model with features from multiple VEs can improve the performance on a target task, and how they are combined. We exhaustively experiment with three popular VEs on six downstream V+L tasks and analyze the attention and VE-dropout patterns. Our analyses suggest that diverse VEs complement each other, resulting in improved downstream V+L task performance, where the improvements are not due to simple ensemble effects (i.e. the performance does not always improve when increasing the number of encoders). We demonstrate that future VEs, which are not repurposed, but explicitly designed for V+L tasks, have the potential of improving performance on the target V+L tasks.
Anthology ID:
2023.repl4nlp-1.9
Volume:
Proceedings of the 8th Workshop on Representation Learning for NLP (RepL4NLP 2023)
Month:
July
Year:
2023
Address:
Toronto, Canada
Editors:
Burcu Can, Maximilian Mozes, Samuel Cahyawijaya, Naomi Saphra, Nora Kassner, Shauli Ravfogel, Abhilasha Ravichander, Chen Zhao, Isabelle Augenstein, Anna Rogers, Kyunghyun Cho, Edward Grefenstette, Lena Voita
Venue:
RepL4NLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
97–117
Language:
URL:
https://aclanthology.org/2023.repl4nlp-1.9
DOI:
10.18653/v1/2023.repl4nlp-1.9
Bibkey:
Cite (ACL):
Gregor Geigle, Chen Liu, Jonas Pfeiffer, and Iryna Gurevych. 2023. One does not fit all! On the Complementarity of Vision Encoders for Vision and Language Tasks. In Proceedings of the 8th Workshop on Representation Learning for NLP (RepL4NLP 2023), pages 97–117, Toronto, Canada. Association for Computational Linguistics.
Cite (Informal):
One does not fit all! On the Complementarity of Vision Encoders for Vision and Language Tasks (Geigle et al., RepL4NLP 2023)
Copy Citation:
PDF:
https://aclanthology.org/2023.repl4nlp-1.9.pdf