๐ฉ ๐ฎ โ e โ g โ m โ e โ n โ t {\mathcal{N}}^{\mathcal{S}egment} : Label-specific Deformations for Remote Sensing Image Segmentation
Yechan Kim , Member , IEEE, DongHo Yoon, SooYeon Kim, Moongu Jeon , Senior Member , IEEE ยท 2025
Read ๐ฉ ๐ฎ โ e โ g โ m โ e โ n โ t {\mathcal{N}}^{\mathcal{S}egment} : Label-specific Deformations for Remote Sensing Image Segmentation by Yechan Kim , Member , IEEE, DongHo Yoon, SooYeon Kim, Moongu Jeon , Senior Member , IEEE (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-19634.txt.
Description
Yechan Kim, DongHo Yoon, SooYeon Kim, and Moongu Jeon tackle mislabeled training data for satellite image mapping in this 2025 paper. Maps that sort land into classes depend on hand labels, but fuzzy borders, mixed pixels, and shadows hide many errors. Labeled satellite data is also scarce, since labeling is costly.
Their method, NSegment, warps only the labels during training, stretching them by varied amounts each round. The trick teaches models to tolerate label sloppiness instead of memorizing it. Unlike selection or correction schemes, it adds little training time or complexity. Tests showed better mapping scores across several leading models. The authors offer the method as a simple fix for noisy label sets.
Catalog record
| CREATOR | Yechan Kim , Member , IEEE, DongHo Yoon, SooYeon Kim, Moongu Jeon , Senior Member , IEEE |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-19634