Summary of Empowering Llms with Pseudo-untrimmed Videos For Audio-visual Temporal Understanding, by Yunlong Tang et al.
Empowering LLMs with Pseudo-Untrimmed Videos for Audio-Visual Temporal Understanding
by Yunlong Tang, Daiki Shimada, Jing Bi, Mingqian Feng, Hang Hua, Chenliang Xu
First submitted to arxiv on: 24 Mar 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper introduces PU-VALOR, a large-scale audio-visual dataset for training multimodal language models to understand the temporal alignment of audio-visual events in videos. The authors fine-tune a model called AVicuna on this dataset and demonstrate its ability to temporally localize audio-visual events and perform well on various tasks such as video QA, audio-visual QA, and event dense localization. By leveraging multimodal large language models (LLMs) and temporal annotations from well-annotated datasets like dense video captioning datasets, the authors show that AVicuna can align audio-visual events with temporal intervals and corresponding text tokens. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper creates a new dataset called PU-VALOR to help language models understand videos better. They took an existing dataset called VALOR, which has some annotations, and added more details about specific events in the videos. Then they used this dataset to train a model that can connect what’s happening in a video with words or phrases. The resulting model is good at finding specific moments in videos and understanding what’s being said. |
Keywords
* Artificial intelligence * Alignment
