Summary of Learning to Detour: Shortcut Mitigating Augmentation For Weakly Supervised Semantic Segmentation, by Junehyoung Kwon et al.
Learning to Detour: Shortcut Mitigating Augmentation for Weakly Supervised Semantic Segmentation
by JuneHyoung Kwon, Eunju Lee, Yunsung Cho, YoungBin Kim
First submitted to arxiv on: 28 May 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 proposes a novel approach to weakly supervised semantic segmentation (WSSS), which aims to reduce annotation costs by leveraging weak forms of labels. However, existing WSSS methods often rely on shortcut features that exploit spurious correlations between certain backgrounds and objects, leading to poor generalization performance. The proposed method, called shortcut mitigating augmentation (SMA), generates synthetic representations of object-background combinations not seen in the training data to disentangle object-relevant and background features. SMA-trained classifiers are less dependent on context and focus more on the target object when making predictions. The approach achieves improved semantic segmentation results on PASCAL VOC 2012 and MS COCO 2014 datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research paper solves a problem in computer vision called weakly supervised semantic segmentation. This means it helps machines understand what’s in an image without needing to label every tiny part of the picture. The problem is that machines can trick themselves by relying on easy patterns instead of really understanding what’s going on. To fix this, the authors created a new way to mix up the data so machines don’t rely too much on these shortcuts. This makes them better at recognizing objects in images. The method works well on two popular image datasets. |
Keywords
* Artificial intelligence * Generalization * Semantic segmentation * Supervised
