---
language:
  - "en"
pretty_name: "Processed ABC Dataset for the Tactile MNIST Benchmark Suite"
tags:
  - "Mesh"

license: "other"
task_categories:
  - other
---
# ABC Dataset

Meshes from the [ABC dataset](https://deep-geometry.github.io/abc-dataset/) processed for use in the [Tactile MNIST Benchmark Suite](https://github.com/TimSchneider42/tactile-mnist).

Specifically, meshes were first filtered, including only those that
- Fit in a cuboid of width 300mm and height 300mm
- Are at least 1mm wide and 300mm tall
- Have a volume of at least 1mm³

Moreover, meshes were simplified to have at most 500 faces per cm² of their surface area via [trimesh's `simplify_quadratic_decimation` method](https://trimesh.org/trimesh.scene.scene.html#trimesh.scene.scene.Scene.simplify_quadric_decimation).

Find usage documentation [here](https://github.com/TimSchneider42/tactile-mnist/blob/main/doc/datasets.md#3d-mesh-datasets) and the full dataset [here](https://huggingface.co/datasets/TimSchneider42/abc-dataset).

## License

Note that I am not the creator of the ABC dataset; hence, I hold no copyright over it.

The copyright of the CAD models is owned by their creators. 
For licensing details, see [Onshape Terms of Use 1.g.ii](https://www.onshape.com/en/legal/terms-of-use#your_content).

This version of the dataset does not contain any metadata.
Hence, to identify the creator of a specific model, please cross-reference the model ID with the indices of the [metadata dataset](https://huggingface.co/datasets/TimSchneider42/abc-dataset-meta).