Computational design of cold bent glass facades (2020) ACM Trans. Graph.

Summary:
Trained MDN model and glass panel processed simulation data with Bezier surface approximations.

Instruction:
1. Clone git repository https://github.com/russelmann/cold-glass-acm
2. Unpack chosen archive files into folder cold-glass-acm/data


* File <mdn_model.tar.gz>

Summary: Trained model data.


* File <optimal_panels_data.tar.gz>

Summary: Panel shape optimization data used for model training. In file <optimal_panels_data.parq> each row represents a single optimal panel of thickness 1 mm.

Format: Parquet

Data columns:
	panel_id: unique identificator of a panel (1XXXYYYYYY, XXX - batch id, YYYYYY - panel number)
	K_NN_D: NN control points in range [00 .. 15] with coordinates D in dimensions [0 .. 2] (meters)
	dataset: name of dataset ('train', 'test', 'none')
	total_energy: total mechanical energy of the panel (J)
	panel_id_src: reference to panel with the same boundary (differs from panel_id if panel is obtained during data enrichment phase)
	max_stress: maximal stress value (Pa) [simulation corner triangles are ignored as overly constrained]
	Lp_12_stress: Lp-norm (p = 12) of maximal stress per-triangle two-sided (Pa)