Summary of Incorporating Geo-diverse Knowledge Into Prompting For Increased Geographical Robustness in Object Recognition, by Kyle Buettner et al.
Incorporating Geo-Diverse Knowledge into Prompting for Increased Geographical Robustness in Object Recognition
by Kyle Buettner, Sina Malakouti, Xiang Lorraine Li, Adriana Kovashka
First submitted to arxiv on: 3 Jan 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 Medium Difficulty summary: Existing object recognition models struggle with domain shifts in diverse geographical scenarios due to a lack of robustness. To adapt class representations and improve accuracy, we explore the feasibility of leveraging geographically diverse descriptive knowledge from large language models. Specifically, we examine the effects of integrating this knowledge into zero-shot and learnable soft prompting with CLIP. Our proposed geography knowledge regularization ensures that prompts trained on source sets generalize to target sets. We demonstrate significant accuracy gains (up to +2.8/1.2/1.6) over baselines on DollarStreet while training only on European data, targeting African, Asian, and American datasets. This performance is competitive with few-shot target training, highlighting the potential for geographical robustness in object recognition models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty summary: Right now, computer vision models don’t do well when they’re shown pictures from different parts of the world. They get confused because the things in those pictures are described differently. To fix this, we looked at whether using descriptions from all around the world could help. We used a special type of AI model that knows about lots of different things to see if it could help our computer vision models understand pictures better. What we found was that by using these descriptions, we could make our models more accurate when shown pictures from places they hadn’t seen before. This is important because it means we can use the same models for all sorts of tasks without having to collect lots of new data. |
Keywords
* Artificial intelligence * Few shot * Prompting * Regularization * Zero shot
