Summary of Text Me the Data: Generating Ground Pressure Sequence From Textual Descriptions For Har, by Lala Shakti Swarup Ray et al.
Text me the data: Generating Ground Pressure Sequence from Textual Descriptions for HAR
by Lala Shakti Swarup Ray, Bo Zhou, Sungho Suh, Lars Krupp, Vitor Fortes Rey, Paul Lukowicz
First submitted to arxiv on: 22 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computation and Language (cs.CL); Signal Processing (eess.SP)
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 In human activity recognition (HAR), obtaining high-quality ground truth data is crucial for efficient model training. However, collecting this data through physical sensors can be costly, time-consuming. To address this challenge, we propose Text-to-Pressure (T2P), a deep learning framework that generates extensive ground pressure sequences from textual descriptions of human activities. By combining vector quantization with text-conditioned auto-regressive strategies and discrete latent correlations between text and pressure maps, we achieve high-quality generated pressure sequences with an R-squared value of 0.722, Masked R-squared value of 0.892, and FID score of 1.83. Our framework also enables training HAR models on synthesized data, achieving comparable performance to those trained solely on real data. Furthermore, combining both real and synthesized training data increases the overall macro F1 score by 5.9 percent. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research paper is about creating a new way to make accurate predictions about what people are doing based on how they move. Right now, it’s hard to get this kind of information because it requires expensive and time-consuming sensors. The researchers developed a system that can take written descriptions of human activities and turn them into detailed records of pressure changes on the ground. This helps create better models for recognizing human activities. The new system is shown to be just as good as using real sensor data, and combining both types of data makes the predictions even more accurate. |
Keywords
* Artificial intelligence * Activity recognition * Deep learning * F1 score * Quantization
