Rubber Tree Trunk Segmentation Dataset for Boundary Knowledge Distillation
Description
This dataset supports the manuscript entitled “TrunkFormer Lite: Boundary Knowledge Distillation for Compact Structure-Aware Trunk Segmentation”. The data were collected for boundary-sensitive segmentation of rubber tree trunks in natural plantation scenes. The repository contains RGB images, LabelMe polygon annotations, binary trunk masks generated by merging trunk instances, tree-level train/validation/test split files, configuration files, and scripts for label conversion and boundary-aware mask evaluation. The dataset is intended to support reproducible research on compact semantic segmentation, unified binary-mask evaluation, and teacher-guided boundary knowledge distillation for slender trunk-like targets. The manuscript uses 10,000 RGB images divided at the individual-tree level into training, validation, and test sets of 7,000, 1,500, and 1,500 images, respectively. The target class is rubber tree trunk, with binary mask values representing background and trunk regions. The repository does not include generated experimental result tables or runtime outputs.
Files
Institutions
- Hainan UniversityHainan, Haikou