Access requires accepting the ShapeNet terms of use
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
These meshes are derived from the Breaking Bad dataset (Sellán et al., 2022), whose everyday subset is built from PartNet / ShapeNet models and inherits the ShapeNet terms of use (https://shapenet.org/terms): non-commercial research and educational use only. Requests are reviewed manually.
Log in or Sign Up to review the conditions and access this dataset content.
Breaking Bad STRIVE collision proxies
Normalized complete and fractured meshes from the volume-constrained everyday subset of
Breaking Bad, each paired with a CoACD compound
collision proxy for MuJoCo. Release everyday-vc-coacd-v1.
Contents
| Objects selected | 496 |
| Objects by status | {'complete': 492, 'failed': 4} |
| (object, pattern) pairs | 983 |
| Pairs by piece count | {2: 241, 3: 265, 4: 266, 5: 211} |
| Meshes with proxies | 3888 |
| Repair methods | {'none': 3888} |
| Mean hulls per mesh | 36.53 |
| Hull cap raised to 64 | 2054 |
| Largest hull cap still reached | 1239 |
Layout
manifest.jsonl one row per (object, pattern): pieces with mesh, proxy, mass properties; adjacency
objects.jsonl one row per selected object: status, complete mesh + proxy, normalization, attempts
candidates.jsonl.gz every pattern of the source subset with its piece count
failures.jsonl every failed attempt (stage, error)
provenance.json source checksum, seed, settings, package versions, code revision
checksums.sha256
assets/<Category>/<object_id>/complete.obj | complete.proxy.obj
assets/<Category>/<object_id>/patterns/<pattern>/piece_<i>.obj | piece_<i>.proxy.obj
A *.repaired.obj sits beside a mesh only when the watertight repair changed it; the proxy and
mass properties are computed from the repaired copy in that case.
Conventions
- Frame: metres, Z-up.
world = (native - center) * scale; scale = min(0.34 / max(ex, ey), 0.55 / ez)with the complete mesh's bounding box; every piece uses its object's transform, so pieces are registered to the complete mesh.center(native units) andscaleare inobjects.jsonl. - Proxies: one OBJ per mesh with one
o convex_NNNobject per convex hull. Load each hull as a MuJoCo mesh geom on one body. An attached group of pieces is one body holding all members' hulls. - Mass properties:
volume_m3,center_of_mass, andinertia_unit_density(about the centre of mass, object axes) at density 1 kg/m^3; multiply by the material density. - Adjacency:
[i, j, area_m2]for pieces sharing a fracture wall.
Selection
Patterns (mode_* and fractured_*) with 2–5
non-empty pieces are eligible. Per object, up to 2 patterns are chosen with
seed 20261004: a piece count is drawn uniformly among those available, then that count's most
volume-balanced pattern. balance = worst_piece_balance × evenness, where worst_piece_balance = pieces ×
smallest volume share (near 0 for a chip) and evenness = (1 / Σ share²) / pieces, the effective number of equal
pieces as a fraction (near 1/pieces when one piece dominates); 1 means equal pieces. The second pattern has a
different count when possible. A failed pattern is replaced by the next most balanced one of the same count (at most
3 attempts per slot). candidates.jsonl.gz lists every pattern's
volume_shares and all three scores.
Processing
- Decompress with the upstream algorithm (libigl): weld duplicates and resolve duplicated faces.
- Normalize; validate watertight, consistent winding, positive volume; otherwise repair
(weld → fill holes → voxel fill at 1/72 of the longest extent). A pattern with any piece that needs
voxel_fillrepair is rejected and its slot moves to the next candidate; an object whose complete mesh needs it is dropped. - CoACD 5.0 mm (
real_metric), hulls of at most 128 vertices, resolution 2000, seed 11. The hull cap starts at 32; when a result hits it, CoACD has merged hulls past the 5 mm error to meet it, so the decomposition is repeated with twice the cap, up to 64 (proxy_hull_cap,cap_escalated,cap_attempts).hull_cap_reachedmarks meshes that still hit 64. - Every hull must compile in MuJoCo; degenerate hulls are dropped only when others remain.
Source
Breaking Bad volume_constrained-everyday_compressed.zip from Dataverse (doi:10.5683/SP3/LZNPKB, file 403902,
sha256 23aab6488ed78f2acb472e7ce29e25e6e211c3f4eb8b3b06a415fac4ad5804b5).
Citation
@inproceedings{sellan2022breaking,
title = {Breaking Bad: A Dataset for Geometric Fracture and Reassembly},
author = {Sell{\'a}n, Silvia and Chen, Yun-Chun and Wu, Ziyi and Garg, Animesh and Jacobson, Alec},
booktitle = {NeurIPS Datasets and Benchmarks Track},
year = {2022}
}
@inproceedings{mo2019partnet,
title = {PartNet: A Large-scale Benchmark for Fine-grained and Hierarchical Part-level 3D Object Understanding},
author = {Mo, Kaichun and Zhu, Shilin and Chang, Angel X. and Yi, Li and Tripathi, Subarna and Guibas, Leonidas J. and Su, Hao},
booktitle = {CVPR},
year = {2019}
}
@techreport{chang2015shapenet,
title = {ShapeNet: An Information-Rich 3D Model Repository},
author = {Chang, Angel X. and Funkhouser, Thomas and Guibas, Leonidas and Hanrahan, Pat and Huang, Qixing and Li, Zimo and Savarese, Silvio and Savva, Manolis and Song, Shuran and Su, Hao and Xiao, Jianxiong and Yi, Li and Yu, Fisher},
institution = {Stanford University, Princeton University, Toyota Technological Institute at Chicago},
number = {arXiv:1512.03012},
year = {2015}
}
@article{wei2022coacd,
title = {Approximate Convex Decomposition for 3D Meshes with Collision-Aware Concavity and Tree Search},
author = {Wei, Xinyue and Liu, Minghua and Ling, Zhan and Su, Hao},
journal = {ACM Transactions on Graphics},
volume = {41},
number = {4},
year = {2022}
}
- Downloads last month
- 28