Stereo RGB Dataset for 3D Main-Axis Recovery of Rubber Tree Trunks and 3D-Printed Quasi-Cylindrical Models

This dataset supports research on stereo RGB based 3D main-axis recovery of quasi-cylindrical targets in natural scenes. It contains 4000 synchronized stereo RGB image pairs of rubber tree trunks collected under natural field conditions and controlled validation data from eight 3D-printed quasi-cylindrical models with known reference axes. The dataset includes stereo calibration parameters, target masks, row-profile centerline records, pair-level quality-control results, stereo matching outputs, triangulated 3D centerline points, robust axis fitting results, ablation data, and ground-truth validation metrics. It is designed to support reproducibility of a reliability-aware stereo RGB framework for recovering continuous 3D axes from weakly textured, partially occluded, and structurally inconsistent stereo observations.