Summary of On Train-test Class Overlap and Detection For Image Retrieval, by Chull Hwan Song et al.
On Train-Test Class Overlap and Detection for Image Retrieval
by Chull Hwan Song, Jooyoung Yoon, Taebaek Hwang, Shunghyun Choi, Yeong Hyeon Gu, Yannis Avrithis
First submitted to arxiv on: 1 Apr 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 The paper revisits the popular Google Landmarks v2 training set by removing class overlap with Revisited Oxford and Paris, a common evaluation set. This change has a significant impact on performance, with varying effects across different methods. To address this issue, the authors introduce CiDeR, an end-to-end pipeline for object detection and global image representation. They demonstrate improved performance on both existing training sets and the new RGLDv2-clean dataset. The study highlights the importance of considering class overlap when evaluating image retrieval models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you’re trying to find a specific picture online. Usually, this is done by searching for keywords or using tags to describe what’s in the picture. But what if there are too many pictures with similar descriptions? This can make it harder to find the one you want. The authors of this paper looked at how we train and test image retrieval models to see if removing duplicate labels helps. They found that making these changes actually makes things worse, but they also came up with a new way to do things called CiDeR. This method does object detection and gets an overall picture representation all in one step. It works better than previous methods on both old and new datasets. |
Keywords
» Artificial intelligence » Object detection
