SpecFormer: Full-Spectrum Optical Thin-Film Dataset for Inverse Design

11,000,000 samples for training and evaluating optical thin-film models. This dataset, created by HongGongLi, provides compact uint16 spectral targets alongside structural labels and fabrication constraints. It was last updated on Hugging Face in May 2026.

Use Cases

• Inverse design of optical multilayer thin films based on structural labels and fabrication constraints
• Spectrum-conditioned modeling tasks using the 774-dimensional spectral targets
• Training and validation of machine learning models for optical property prediction
• Benchmarking algorithms for material design based on VIS-R, VIS-T, NIR-R, and NIR-T data

Strengths

• Contains 10,000,000 training samples, providing a substantial base for model development
• Includes 500,000 validation and 500,000 test samples for robust evaluation
• Each sample stores a 774-dimensional spectral target, offering detailed optical property data

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Description metadata is limited; actual data quality requires manual inspection after download

Provenance

Source

HongGongLi on Hugging Face

Collection Method

Parquet export of the SpecFormer dataset; original data generation method is not specified.

Time Range

null

Freshness

Last updated 2026-05-25 13:47:49; freshness should be verified

Geography

null