DiffPlantCT: a training-free, annotation-free approach to cross-species plant CT image segmentation
Weizhen Liu, Yihao Fan, Haowei Zhao, Chang Chi, Yihong Wu, Hao Lu, Weijuan Hu, Xiaohui Yuan · 2026
Read DiffPlantCT: a training-free, annotation-free approach to cross-species plant CT image segmentation by Weizhen Liu, Yihao Fan, Haowei Zhao, Chang Chi, Yihong Wu, Hao Lu, Weijuan Hu, Xiaohui Yuan (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/PMC13491703.txt.
Description
Weizhen Liu and seven co-authors introduce in 2026 a method for outlining plant structures inside CT scans without any training data or hand labeling. Standard deep learning approaches demand large sets of painstakingly labeled images for each new species, which slows cross-species research. Their system, DiffPlantCT, borrows visual features from a pretrained image generation model and combines them with the scan's own gray levels.
Each two-dimensional slice is segmented by unsupervised clustering, and the slices are then stacked back into a three-dimensional result. Tests ran on walnut fruit and barley spike datasets plus self-collected wheat and rice material. The method beat the best previous unsupervised approach by a wide margin on overlap scores. The authors present it as the first proof that plant CT segmentation can work with neither annotations nor training.
Catalog record
| CREATOR | Weizhen Liu, Yihao Fan, Haowei Zhao, Chang Chi, Yihong Wu, Hao Lu, Weijuan Hu, Xiaohui Yuan |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63608 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |