LaForge10k: Synthetic Herbarium Specimen Images with Semantic Segmentation Annotations

Procedurally generated synthetic training data for semantic segmentation of herbarium specimen images. class IDs are intentionally spaced out such that they can be visibly differentiatedclass_ids = {'other':0, 'paper':5, 'crc':50, 'scale':100, 'woody':150, 'label':200, 'leaf':250}本数据集为面向植物标本图像语义分割（semantic segmentation）任务的程序化生成合成训练数据集。为使各分类ID可通过视觉直观区分，各类别ID被刻意设置为间隔分布： class_ids = {'other':0, 'paper':5, 'crc':50, 'scale':100, 'woody':150, 'label':200, 'leaf':250}