Summary of Motion Guided Token Compression For Efficient Masked Video Modeling, by Yukun Feng et al.
Motion Guided Token Compression for Efficient Masked Video Modeling
by Yukun Feng, Yangming Shi, Fengze Liu, Tan Yan
First submitted to arxiv on: 10 Jan 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 Recent advancements in Transformers have significantly improved video comprehension capabilities. However, the O(N^2) computation complexity associated with attention mechanisms becomes a substantial challenge when dealing with high-dimensional videos and increasing frames per second (FPS). This is because increased FPS rates introduce redundancy and exacerbate existing computational limitations. To address this issue, we propose Motion Guided Token Compression (MGTC), a novel approach that empowers Transformer models to utilize a smaller yet more representative set of tokens for comprehensive video representation. MGTC achieves substantial reductions in computational burden while remaining adaptable to increased FPS rates. Our experiments on Kinetics-400, UCF101, and HMDB51 demonstrate that elevating the FPS rate results in significant top-1 accuracy score improvements, with a maximum gain of 4.0. By implementing MGTC with a masking ratio of 25%, we further augment accuracy by 0.1 while reducing computational costs by over 31% on Kinetics-400. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about using special computer models called Transformers to understand videos better. The problem is that these models get very slow when trying to analyze lots of video frames quickly. To solve this, the researchers created a new way to make the model work more efficiently by selecting only the most important parts of each frame. They tested this method on several famous video recognition datasets and found that it can improve how well the model understands videos while also using less computer power. |
Keywords
» Artificial intelligence » Attention » Token » Transformer
