README.md

---
pretty_name: GeometricBreaks
language:
- en
size_categories:
- 10K<n<100K
tags:
- 3d
---
DATASET DESCRIPTION:

The Geometric Breaks dataset consists of 3D models of 25,249 broken object models and their repair parts. 
The broken models are generated by subtracting fracturing shapes from a total of 22,165 3D complete object models. 
Complete object models have been acquired from the [Google Scanned Object (GSO) dataset](https://app.gazebosim.org/GoogleResearch/fuel/collections/Scanned%20Objects%20by%20Google%20Research), and 8 classes in 
the [ShapeNetCore.v2 dataset](https://huggingface.co/datasets/ShapeNet/ShapeNetCore). 
To generate the dataset, we have waterproofed complete meshes, normalized them to a [-0.5,0.5]^3 cube, and 
generated broken shapes and repair counteparts by conducting Boolean subtraction with a fracturing shape from the 
waterproofed complete mesh. Fracturing shapes come from one of three geometric primitives: cube, icosphere, and subivided icosphere, 
randomly translated and rotated to generate breaks over various parts of the original model. 
Vertices on the primitive surface are randomly perturbed to simulate roughness. 
Since the subtraction process generates multiple fragments, the dataset also contains individual 3D models of the fragmented parts. 
Models from following ShapeNetCore.v2 classes are used: 
Jar (03593526), Bottle (02876657), Mug (03797390), Airplane (02691156), Chair (03001627), Car (02958343), Sofa (04256520), Table (04379243).

DATASET STRUCTURE:
The dataset is shared as the following ZIP files:
- kitchenware.zip: jar, bottle and mug objects
- furniture.zip: table, chair and sofa objects
- vehicles.zip: airplane and car objects
- GSO.zip: GSO objects

Each zip file when unzipped generates the folders corresponding to its object classes. 
For GSO.zip, a folder called GSO is created. 
For the remaining zip files, folders are created labeled by the the ShapeNetCore IDs of the 8 classes used in the dataset. 
Within each class folder, object models are stored in the subfolder hierarchy '[classID]/[objectID]/models/'. 
The value of '[classID]' corresponds to the ShapeNetCore ID or 'GSO'. 
The value of '[objectID]' corresponds to the ShapeNetCore object model ID for ShapeNetCore models and 
to the object model descriptor name with underscores as used to name object models in GSO. 
The use of the same ShapeNetCore class IDs, and ShapeNetCore or GSO object IDs enables 
one-to-one cross-referencing with the original ShapeNetCore and GSO datasets.

Within each '[classID]/[objectID]/models' subfolder, models are stored as triangular meshes in the [PLY file format](https://gamma.cs.unc.edu/POWERPLANT/papers/ply.pdf) as follows: 
- The waterproofed version of the original model is provided in the file 'model_c.ply', where 'c' stands for complete. 
- Fragments generated from the fractured (i.e. 'broken') model mesh are stored as 'model_r_[N]\_fragment\_[M].ply', where,
'b' stands for 'broken', N E {0,1,...,N_max-1} represents the index for the N+1th fractured mesh,
and M E {0,1,...} represents the index for the M+1th fragment in decreasing order of volume.
N_max is 3 for jars and bottles, 10 for mugs, and 1 for the rest of the classes.
The fragment 'model_b_[N]\_fragment\_0.ply' represents the broken mesh fragment with the largest volume.
- Fragments generated from the repair mesh are stored as 'model_r_[N]\_fragment\_[M].ply', 
where, 'r' stands for 'repair' and M E {0,1,...} represents the index for the M+1th repair mesh fragment. 
The number of repair mesh fragments typically differs from the number of broken mesh fragments.
 
In the interest of storage, the original broken and repair meshes generated from subtraction (i.e., the meshes prior to 
fragmenting connecting components) are not shared. However, they can be generated as a union or merge of the corresponding mesh's fragments.
The helper script 'merge_fragments.py' has been shared to perform merging.