Summary of Improving Weakly-supervised Object Localization Using Adversarial Erasing and Pseudo Label, by Byeongkeun Kang and Sinhae Cha and Yeejin Lee
Improving Weakly-Supervised Object Localization Using Adversarial Erasing and Pseudo Label
by Byeongkeun Kang, Sinhae Cha, Yeejin Lee
First submitted to arxiv on: 15 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 A framework for weakly-supervised object localization is introduced, which trains a neural network to predict both object class and location using only images and image-level class labels. The approach consists of a shared feature extractor, classifier, and localizer, with the localizer predicting pixel-level class probabilities and the classifier predicting the object class at the image level. To improve localization accuracy, novel losses are designed that utilize adversarially erased foreground features and feature maps, as well as pseudo labels to suppress activation values in the background while increasing them in the foreground. The proposed method is evaluated on three publicly available datasets (ILSVRC-2012, CUB-200-2011, and PASCAL VOC 2012) using two backbone networks (MobileNetV1 and InceptionV3), demonstrating state-of-the-art performance across all evaluated metrics. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper shows how to train a computer vision model without needing lots of labeled data. It’s like trying to find a specific person in a crowd by looking at the overall vibe or color of the people around them, rather than individually labeling each person. The method uses a combination of techniques to help the model focus on the right things and ignore distractions. It’s tested on three different datasets and outperforms other methods. |
Keywords
» Artificial intelligence » Neural network » Supervised
