Summary of Towards Realistic Incremental Scenario in Class Incremental Semantic Segmentation, by Jihwan Kwak et al.
Towards Realistic Incremental Scenario in Class Incremental Semantic Segmentation
by Jihwan Kwak, Sungmin Cha, Taesup Moon
First submitted to arxiv on: 16 May 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
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 approach to incremental semantic segmentation is proposed, addressing the unrealistic “overlapped” scenario where images reappear with different labels, which can lead to biased results. The authors identify flaws in two common techniques and propose a practical “partitioned” scenario that divides datasets into subsets for each class. Additionally, they address code implementation issues related to exemplar memory retrieval and introduce a simple yet competitive baseline, MiB-AugM, achieving state-of-the-art results across multiple tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper talks about how we learn new things from images, but the way we do it now is not very realistic. Imagine if you saw the same picture of a dog twice, but the second time the picture showed a cat instead! That’s kind of what happens when we use a common method called Continuous Incremental Semantic Segmentation (CISS). The authors found that this method can lead to unfair results for certain techniques. To fix this, they suggest dividing the images into groups based on what’s in them, like dogs and cats. They also fixed some coding problems and created a new way to learn from these images called MiB-AugM, which is really good at recognizing changes in the background. |
Keywords
» Artificial intelligence » Semantic segmentation
