Align and Attend: Multimodal Summarization with Dual Contrastive Losses
CVPR 2023

1 University of Maryland, College Park   2 Carnegie Mellon University   3 Adobe Research

Video

Abstract

The goal of multimodal summarization is to extract the most important information from different modalities to form summaries. Unlike unimodal summarization, the multimodal summarization task explicitly leverages cross-modal information to help generate more reliable and high-quality summaries. However, existing methods fail to leverage the temporal correspondence between different modalities and ignore the intrinsic correlation between different samples. To address this issue, we introduce Align and Attend Multimodal Summarization (A2Summ), a unified multimodal transformer-based model which can effectively align and attend the multimodal input. In addition, we propose two novel contrastive losses to model both inter-sample and intra-sample correlations. Extensive experiments on two standard video summarization datasets (TVSum and SumMe) and two multimodal summarization datasets (Daily Mail and CNN) demonstrate the superiority of A2Summ, achieving state-of-the-art performances on all datasets. Moreover, we collected a large-scale multimodal summarization dataset BLiSS, which contains livestream videos and transcribed texts with annotated summaries.

Model overview

Our A2Summ is a unified noval multimodal transformer-based model which can effectively align and attend the multimodal input.

  • The Alignment-Guided Self-Attention module exploits the time correspondence between video and text modalities.
  • The Inter-Sample Contrastive Loss utilizes the intrinsic relationships between each input video and text pair.
  • The Intra-Sample Contrastive Loss models more fine-grained information across modalities.

Qualitative




BLiSS Dataset

Statistics comparisons

Example

Note that the extractive text summary is formed by the key sentences, where the ground-truth keywords are marked with blue color.

Citation

@inproceedings{he2023a2summ,
  title = {Align and Attend: Multimodal Summarization with Dual Contrastive Losses},
  author = {He, Bo and Wang, Jun and Qiu, Jielin and Bui, Trung and Shrivastava, Abhinav and Wang, Zhaowen},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2023},
}

The website template was borrowed from Ben Mildenhall.