Summary of Cood: Concept-based Zero-shot Ood Detection, by Zhendong Liu et al.
COOD: Concept-based Zero-shot OOD Detection
by Zhendong Liu, Yi Nian, Henry Peng Zou, Li Li, Xiyang Hu, Yue Zhao
First submitted to arxiv on: 15 Nov 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: The paper presents a novel approach to detecting out-of-distribution (OOD) samples in complex multi-label settings without requiring extensive retraining. Current methods struggle to capture the intricate relationships between labels, often relying on large amounts of training data and failing to generalize to unseen combinations. To address this challenge, the authors introduce COOD, a zero-shot multi-label OOD detection framework that leverages pre-trained vision-language models and a concept-based label expansion strategy. The approach enriches the semantic space with both positive and negative concepts for each label, modeling complex dependencies and precisely differentiating OOD samples without additional training. Experimental results demonstrate significant performance improvements over existing methods, achieving average AUROC of approximately 95% on VOC and COCO datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty summary: This paper is about finding things that don’t belong in a group of labeled pictures or text. Right now, computers have trouble doing this when there are multiple labels involved, like when a picture has both animals and plants in it. The authors create a new way to detect these unusual items without needing lots more training data. They use special language models that understand words and images together, and they add extra information about each label to help the computer make better decisions. This method works really well, beating other methods by a lot on two different datasets. |
Keywords
* Artificial intelligence * Zero shot
