Cut to the Mix: Simple Data Augmentation Outperforms Elaborate Ones in Limited Organ Segmentation Datasets
Chang Liu, Fuxin Fan, Annette Schwarz, Andreas Maier · 2026
Read Cut to the Mix: Simple Data Augmentation Outperforms Elaborate Ones in Limited Organ Segmentation Datasets by Chang Liu, Fuxin Fan, Annette Schwarz, Andreas Maier (2026) 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-2602-03555.txt.
Description
Segmenting organs in medical scans needs many hand-labeled images, which clinics rarely have. Chang Liu and 3 co-authors compare data augmentation strategies for that low-data setting in a 2026 paper. They test four inter-image methods, CutMix, CarveMix, ObjectAug, and AnatoMix, which build new training images by mixing separate patients' scans.
On two organ segmentation datasets, three of the four methods improve the average Dice score over a strong nnU-Net baseline with no augmentation. The simple CutMix wins, gaining nearly five points, even though its pasted composites look intuitively wrong to a human eye. Adding traditional within-image augmentation improves results further. The authors release their implementation as a benchmark for future work.
Catalog record
| CREATOR | Chang Liu, Fuxin Fan, Annette Schwarz, Andreas Maier |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 19 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03555