ZCTG: A Zero-Shot Framework for Automatic Video Chaptering and Title Generation
Keywords: Video Chapter Generation, Large Language Models, Scene Graph
TL;DR: A zero-shot framework for automatic video chapter and title generation using only videos as input to facilitate easy navigation and retrieval of video content.
Abstract: In the vast landscape of video content, breaking down lengthy videos into chapters accompanied by concise, descriptive titles greatly enhances searchability and retrieval efficiency. While recent advancements in this field often incorporate multiple data modalities along with human-annotated chapter titles, access to such data, like speech transcripts or audio, is not always guaranteed. Moreover, the manual annotation of chapter titles is expensive and time-consuming. To address these challenges, we introduce ZCTG, a novel and unified zero-shot framework designed to generate video chapters and their concise titles for untrimmed videos. ZCTG utilizes the combined capabilities of scene graphs and Large Language Models (LLMs). The advantages of ZCTG are three-fold: 1) offers practical utility, relying solely on video data; 2) eliminates the need for detailed chapter title supervision; 3) exhibits excellent generalization capabilities in a completely zero-shot setting, without any training needed. We conduct an extensive evaluation on VidChapters-7M and GTEA datasets, which include videos of varying duration and domains, to demonstrate the efficacy of our proposed framework.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 5274
Loading