Introduce

The FragTag3D dataset is designed for shattered 3D models and comes annotated with overlapping area information between fragments. Based on the Thingy10k dataset, it employs Blender's fracture_modifier plugin to create shattered 3D models. This dataset features a variety of thick-walled shattered 3D models, with corresponding fracture surfaces color-coded to facilitate automatic assembly and machine learning applications. Furthermore, we have extended these overlapping area annotations to the 3D Puzzle dataset from TU Wien (Vienna University of Technology). Available in open 3D formats and accompanied by comprehensive metadata, this dataset is of significant importance for researchers and practitioners in fields such as archaeological artifact assembly, machine learning, and point cloud registration.

Contents

As of the publication of this paper, the FragTag3D dataset contains a total of 294 fragmented objects. Of these, 292 are geometrically simulated fragmented objects sourced from Thingi10k, and 2 are from the 3D Puzzle fragmented object dataset. The total size of the dataset is approximately 30GB, and the file format is PLY. All fragmented objects are stored in their original poses; the annotations are represented by colors in the format (r, g, b) and have not been normalized. The annotation information is embedded within the PLY files. FragTag3D is a continuously growing dataset, with the number of annotated fragmented objects still expanding. We aim to include a total of 500 groups of fragmented objects from Thingi10k and also plan to select and annotate more data from other sources. The annotation process is ongoing and long-term.

Object Demo