umangijain/material-magic-wand

Material Magic Wand

Dataset for Material Magic Wand: Material-Aware Grouping of 3D Parts in Untextured Meshes (CVPR 2026).

[Project Page] [Paper] [Code]

Evaluation Testbench

The testbench consists of 100 untextured meshes from Objaverse with human-refined material grouping labels and multi-view renderings.

test-bench/
├── objaverse_uids.json         # 100 Objaverse asset UIDs
├── labels/
│   └── {uid}/
│       └── {material_id}_final.json
├── val_renders/                # Deduplicated renders (used in the paper)
│   ├── metadata.json
│   └── images/
│       └── {uid}/
└── val_renders_full/           # Full renders (all parts, all meshes)
    ├── metadata.json
    └── images/
        └── {uid}/

Objaverse UIDs

test-bench/objaverse_uids.json lists 100 Objaverse asset UIDs. Download the meshes via the Objaverse Python API:

import objaverse, json

with open("test-bench/objaverse_uids.json") as f:
    uids = json.load(f)

objects = objaverse.load_objects(uids=uids)

Labels

test-bench/labels/{uid}/ contains one JSON per material group, with the following fields:

Field	Description
material_id	Integer index for the material group
original_selection	Part indices from the original Objaverse labeling (may be absent)
primary_query	The representative part index for this group
final_selection	Refined list of part indices belonging to this material group

Renders

Each mesh part is rendered as a PNG at three resolutions: full, medium, and small. Images are named {uid}_{part_id}_{part_id}_{view}_{size}.png.

Deduplicated renders (val_renders/): Parts that are geometric duplicates of each other share a single rendering. This is the variant used in the paper. val_renders/metadata.json maps each unique (rendered) part to the list of duplicate part indices it represents.

Full renders (val_renders_full/): Renders for every part of every mesh, without deduplication.

Citation

@inproceedings{materialmagicwand2026,
  title={Material Magic Wand: Material-Aware Grouping of 3D Parts in Untextured Meshes},
  author={Jain, Umangi and Kim, Vladimir and Gadelha, Matheus and Gilitschenski, Igor and Chen, Zhiqin},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2026}
}