Summary of Exclusive Style Removal For Cross Domain Novel Class Discovery, by Yicheng Wang et al.
Exclusive Style Removal for Cross Domain Novel Class Discovery
by Yicheng Wang, Feng Liu, Junmin Liu, Kai Sun
First submitted to arxiv on: 26 Jun 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 novel class discovery (NCD) task is typically used to cluster unseen classes in an unlabeled set based on labeled data within the same domain. However, existing methods may struggle when novel classes are sampled from a different distribution than the labeled ones. This paper explores and establishes the solvability of NCD in cross-domain settings, requiring style information removal. Theoretical analysis leads to an exclusive style removal module that extracts distinctive style features, facilitating inference. This module can be easily integrated with other NCD methods, serving as a plug-in to improve performance on novel classes with different distributions. A fair benchmark is also built to account for the influence of backbones and pre-training strategies on NCD method performance. Extensive experiments on three datasets demonstrate the effectiveness of the proposed module. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Novel class discovery (NCD) is a way to group new things we’ve never seen before into categories. Right now, this task can be tricky when these new things come from a different place than what we’re familiar with. This paper looks at how we can make NCD work even better in situations like this. It creates a special tool that helps take out extra information that might confuse the process. This tool is easy to use with other methods, and it makes the results better when dealing with new things from different places. The paper also sets up a fair way to test these methods and see which ones work best. |
Keywords
» Artificial intelligence » Inference
