Summary of Llm Meets Vision-language Models For Zero-shot One-class Classification, by Yassir Bendou et al.
LLM meets Vision-Language Models for Zero-Shot One-Class Classification
by Yassir Bendou, Giulia Lioi, Bastien Pasdeloup, Lukas Mauch, Ghouthi Boukli Hacene, Fabien Cardinaux, Vincent Gripon
First submitted to arxiv on: 31 Mar 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 tackles the problem of zero-shot one-class visual classification, where a model must distinguish between positive and negative query samples without having any examples from the target class. The proposed method consists of two steps: first, querying large language models for visually confusing objects, and then relying on vision-language pre-trained models (e.g., CLIP) to perform classification. To demonstrate the effectiveness of this approach, the authors adapt large-scale vision benchmarks and show that their method outperforms adapted off-the-shelf alternatives in this setting. Specifically, they propose a realistic benchmark where negative query samples are drawn from the same original dataset as positive ones, including a granularity-controlled version of iNaturalist. This paper presents a novel solution for discriminating between a single category and semantically related ones using only their label. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine trying to identify a specific type of animal without having any pictures of that animal, but you have lots of pictures of other animals that look similar. This is the problem this paper solves. The researchers developed a way to use language models and vision models together to identify one category of objects (like birds) from others (like fish). They tested their approach on big datasets and showed it works better than using other existing methods. The goal is to be able to recognize something without having any examples, just by knowing what the thing is called. |
Keywords
» Artificial intelligence » Classification » Zero shot
