---
license: mit
tags:
  - electromagnetics
  - nanophotonics
  - jcmwave
  - optical-constants
  - parquet
configs:
- config_name: metadata
  default: true
  data_files:
  - split: train
    path: metadata/train-*
- config_name: validation
  data_files:
  - split: train
    path: validation/train-*
---

# squiggles (metadata-fix)

OC-map FEM rebuild at **35 pixels per wavelength**, with corrected geometries,
Helmholtz residuals, and the **resolved JCMsuite `.jcm` / `.jcmp` files** used
for each solve.

## Configs

### `metadata` (default)

One row per structure folder (`sample_XXXX`). Geometry comes from published
optical-constant maps (not the old nested-interface metadata).

### `validation`

One row per FEM incidence (`theta` in `{0, 45}`). Self-contained pixel map:

```text
pitch_nm = wavelength_nm / pixels_per_wavelength
x_nm = x0_nm + ix * pitch_nm
y_nm = y0_nm + iy * pitch_nm
```

Also includes nested OC/E arrays, Helmholtz scalar metrics (s-pol staggered FD),
and binary blobs of the generated project files:

- `layout` <- `layout.jcm`
- `materials` <- `materials.jcm`
- `sources` <- `sources.jcm`
- `project` <- `project.jcmp` (JCMsuite expanded project)
- `grid` <- `grid.jcm`

## Load

```python
from datasets import load_dataset
meta = load_dataset("als-rixs/latent-image-training", "metadata", split="train", revision="metadata-fix")
val = load_dataset("als-rixs/latent-image-training", "validation", split="train", revision="metadata-fix")
row = val[0]
layout_jcm = row["layout"]  # bytes
```