English

COSA: Concatenated Sample Pretrained Vision-Language Foundation Model

Computer Vision and Pattern Recognition 2023-06-16 v1 Artificial Intelligence Computation and Language Machine Learning Multimedia

Abstract

Due to the limited scale and quality of video-text training corpus, most vision-language foundation models employ image-text datasets for pretraining and primarily focus on modeling visually semantic representations while disregarding temporal semantic representations and correlations. To address this issue, we propose COSA, a COncatenated SAmple pretrained vision-language foundation model. COSA jointly models visual contents and event-level temporal cues using only image-text corpora. We achieve this by sequentially concatenating multiple image-text pairs as inputs for pretraining. This transformation effectively converts existing image-text corpora into a pseudo long-form video-paragraph corpus, enabling richer scene transformations and explicit event-description correspondence. Extensive experiments demonstrate that COSA consistently improves performance across a broad range of downstream tasks, including long-form/short-form video-text tasks and image-text tasks such as retrieval, captioning, and question answering. Notably, COSA achieves state-of-the-art results on various competitive benchmarks. Code and model are released at https://github.com/TXH-mercury/COSA.

Keywords

Cite

@article{arxiv.2306.09085,
  title  = {COSA: Concatenated Sample Pretrained Vision-Language Foundation Model},
  author = {Sihan Chen and Xingjian He and Handong Li and Xiaojie Jin and Jiashi Feng and Jing Liu},
  journal= {arXiv preprint arXiv:2306.09085},
  year   = {2023}
}
R2 v1 2026-06-28T11:05:53.989Z