README.md

---
license: mit
pretty_name: MTO 2D v0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: val
    path: data/val-*
  - split: test
    path: data/test-*
---

# MTO2D v0

EngiBench-compatible MTO2D dataset published as
`IDEALLab/mto_2d_v0`. It contains 5,666 converged `(400, 200)`
half-domain designs with train/val/test splits of
4,249/283/1,134.

Objectives: `mean_temperature` and `power_dissipation` are historical source-optimization
labels logged before the final MMA/Heaviside update. The stored `app/200/gamma`
design is the subsequent post-update density field.

For the EngiBench problem, a frozen simulator uses the following physics:
`q=0.01`, `alphaMax=5025200`,
`Heaviside=59.8`, with design updates disabled.

`optimal_design` is a flattened float list of length
80,000, reshaping to `(400, 200)` and symmetrical about the vertical axis so as to yield a `(400, 400)` heat sink.

## Citation

```bibtex
@article{drake2026quantize,
  title={To Quantize or Not to Quantize: Effects on Generative Models for Topology Optimization Problems},
  author={Drake, Arthur and Chen, Qiuyi and Wang, Jun and Nejat, Ardalan and Guest, James K and Fuge, Mark},
  journal={Journal of Mechanical Design},
  volume={148},
  number={10},
  pages={101704},
  year={2026},
  publisher={American Society of Mechanical Engineers}
}
```