Automated shape repair approaches currently lack access to datasets that describe real-world damage geometry. We present Fantastic Breaks (and Where to Find Them: https://terascale-all-sensing-research-studio.github.io/FantasticBreaks), a dataset containing scanned, waterproofed, and cleaned 3D meshes for 78 broken objects, paired and geometrically aligned with complete counterparts. Fantastic Breaks contains class and material labels, synthetic proxies of repair parts that join to broken meshes to generate complete meshes, and manually annotated fracture boundaries. Through a detailed analysis of fracture geometry, we reveal differences between Fantastic Breaks and datasets of synthetically fractured objects generated using geometric and physics-based methods. We show experimental results of shape repair with Fantastic Breaks using multiple learning-based approaches pre-trained using a synthetic dataset and re-trained using a subset of Fantastic Breaks.

翻译：自动形状修复方法目前缺乏描述真实损伤几何形状的数据集。我们提出了“现实中的碎片（及其寻找方法：https://terascale-all-sensing-research-studio.github.io/FantasticBreaks）”，该数据集包含78个破损物体的扫描、防水处理及清理后的三维网格，并与其完整对应物配对且实现几何对齐。该数据集包含类别与材料标签、修复部件的合成代理（可连接至破损网格以生成完整网格），以及人工标注的断裂边界。通过对断裂几何形状的详细分析，我们揭示了该数据集与使用几何及物理方法生成的合成断裂物体数据集之间的差异。我们展示了使用该数据集进行形状修复的实验结果，其中采用了多种基于学习的方法——这些方法先使用合成数据集进行预训练，再使用该数据集的一个子集进行再训练。