Summary of When Text and Images Don’t Mix: Bias-correcting Language-image Similarity Scores For Anomaly Detection, by Adam Goodge et al.
When Text and Images Don’t Mix: Bias-Correcting Language-Image Similarity Scores for Anomaly Detection
by Adam Goodge, Bryan Hooi, Wee Siong Ng
First submitted to arxiv on: 24 Jul 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 This paper explores the limitations of Contrastive Language-Image Pre-training (CLIP) models, which excel in various downstream tasks by aligning image and text input embeddings. The study reveals that despite being trained for contrast, the text inputs’ embeddings unexpectedly cluster together, away from image embeddings, inducing a “similarity bias” that affects anomaly detection performance. To mitigate this issue, the authors propose BLISS, a novel methodology that directly addresses the similarity bias by incorporating an external set of text inputs. This approach is simple, requiring no strong inductive biases or expensive training processes, and outperforms baseline methods on benchmark image datasets, even when access to normal data is limited. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how well machines can learn from pictures and words. They found that a popular way to do this called CLIP doesn’t always work as expected. Despite being designed to make the two very different things – images and text – similar, it actually makes them very different instead! This causes problems when trying to find unusual patterns or things that don’t belong. The researchers came up with a new idea called BLISS that helps fix this issue. It’s easy to use and doesn’t need lots of training data, making it useful for real-world applications. |
Keywords
» Artificial intelligence » Anomaly detection
