PlantC2USeg: Cross-Scale Consistent Pre-Training for Few-Shot Unified Plant Point Cloud Segmentation

TL;DR

PlantC2USeg is a deep transfer learning framework for plant point cloud segmentation, achieving state-of-the-art results with few-shot learning. It enables scalable plant phenotyping and transferable 3D representation learning. The framework reduces adaptation effort under distribution shifts.

Why It Matters

This research has significant implications for crop breeding and agriculture, enabling precise organ-level analysis with minimal labeled data. It can also be applied to other domains with diverse shape variations. The framework's ability to handle distribution shifts makes it a valuable tool for real-world applications.

Key Findings

• 91.91% semantic IoU on Soybean3D
• 89.78% IoU with 20 labeled samples
• 85.0% category-averaged mIoU on ShapeNet Part