Summary of Exploring Text-to-motion Generation with Human Preference, by Jenny Sheng et al.
Exploring Text-to-Motion Generation with Human Preference
by Jenny Sheng, Matthieu Lin, Andrew Zhao, Kevin Pruvost, Yu-Hui Wen, Yangguang Li, Gao Huang, Yong-Jin Liu
First submitted to arxiv on: 15 Apr 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
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 proposed paper explores preference learning in text-to-motion generation, a task that currently relies on datasets requiring expert labelers with motion capture systems. The authors find that learning from human preference data is more efficient and does not require these resources. Instead, a labeler without expertise can compare two generated motions to determine which one is preferred. This approach has the potential to improve current text-to-motion generative models. The paper annotates 3,528 preference pairs generated by MotionGPT, the first effort to investigate various algorithms for learning from preference data. Experimental results show that preference learning can greatly improve current models. The code and dataset are publicly available. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Text-to-motion generation is a task where computers create motion based on text input. Right now, this process relies on datasets made by experts using special equipment called motion capture systems. Researchers thought it would be better to learn from what people like or don’t like about the generated motions instead. They did this by having someone without expertise compare two generated motions and say which one they preferred. This approach can improve how well computers generate motions based on text. The study looked at 3,528 pairs of these preferred motions to see how different algorithms worked. The results showed that learning from what people like or don’t like has a lot of potential. |
