Summary of Conjugated Semantic Pool Improves Ood Detection with Pre-trained Vision-language Models, by Mengyuan Chen et al.
Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models
by Mengyuan Chen, Junyu Gao, Changsheng Xu
First submitted to arxiv on: 11 Oct 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 A straightforward pipeline for zero-shot out-of-distribution (OOD) detection involves selecting potential OOD labels from an extensive semantic pool and then leveraging a pre-trained vision-language model to perform classification on both in-distribution (ID) and OOD labels. This paper theorizes that enhancing performance requires expanding the semantic pool, increasing the expected probability of selected OOD labels being activated by OOD samples, and ensuring low mutual dependence among these activations. By adopting a larger lexicon or “making up” OOD label candidates not standard class names but beneficial for the process, we can expand the OOD label candidates to meet the requirements and outperform existing works in FPR95 by 7.89%. The proposed conjugated semantic pool (CSP) consisting of modified superclass names serves as a cluster center for samples sharing similar properties across different categories. Codes are available on GitHub. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about creating a new way to detect when something is out of the normal range. They use a combination of words and images to do this, and they think that by using more words and making sure the words are related but not too similar, they can make it work better. They tested their idea and it worked 7.89% better than what others had done before. |
Keywords
» Artificial intelligence » Classification » Language model » Probability » Zero shot
