Summary of Linvt: Empower Your Image-level Large Language Model to Understand Videos, by Lishuai Gao et al.
LinVT: Empower Your Image-level Large Language Model to Understand Videos
by Lishuai Gao, Yujie Zhong, Yingsen Zeng, Haoxian Tan, Dengjie Li, Zheng Zhao
First submitted to arxiv on: 6 Dec 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG); Multimedia (cs.MM)
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 This research paper proposes a novel approach to transform well-trained image-based Large Language Models (LLMs) into video-LLMs by introducing two design principles: linear transformation and representative information condensation. The authors develop a plug-and-play module, Linear Video Tokenizer (LinVT), which enables existing image-LLMs to understand videos. They benchmark LinVT with six recent visual LLMs, showcasing its high compatibility. The resulting LinVT-based LLMs achieve state-of-the-art performance across various video benchmarks, demonstrating the effectiveness of LinVT in multi-modal video understanding. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us develop a new kind of AI assistant that can understand videos. Instead of training an entire new model from scratch, we take existing models that are good at processing images and modify them to work with videos. The authors came up with two important ideas: first, they found a way to keep the original connection between what’s seen in a video and what it means (visual-language alignment). Second, they figured out how to remove unnecessary details from long videos, keeping only the most important information. They created a special tool called Linear Video Tokenizer that lets us use these image-based models for videos. The authors tested their idea with six different models and showed that it works really well, even beating other top-performing models. |
Keywords
» Artificial intelligence » Alignment » Multi modal » Tokenizer
