# Human eye model attribution

**3D Reference Organ for Eye, Female, Left v1.2** (2023), by **Kristen Browne and Heidi Schlehlein**, Human Reference Atlas / HuBMAP. Funded by the National Institutes of Health. Built from the National Library of Medicine Visible Human Female dataset.

- License: [Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/).
- DOI: https://doi.org/10.48539/HBM378.DBLX.423
- Library: https://humanatlas.io/3d-reference-library
- Release metadata and explicit license: https://github.com/hubmapconsortium/ccf-releases/blob/main/v1.4/markdown/ref-organs/3d-vh-f-eye-l.md
- Original GLB: https://raw.githubusercontent.com/hubmapconsortium/ccf-releases/main/v1.4/models/3d-vh-f-eye-l.glb
- Downloaded: 2026-10-09.

The bundled `human-eye.glb` is an unmodified copy of the source GLB, renamed for the application. Runtime display may center, scale, recolor, hide, clip, and separate structures for teaching; those presentation changes are not changes to the source anatomy. Retain this attribution and license link with redistribution. No endorsement by the creators, HRA, HuBMAP, NIH, or NLM is implied.

## Verified asset details

23 individual anatomical meshes; 346,630 triangles; 27,462,204 bytes. Standard glTF 2.0 binary, without required extensions, external images, or decoder dependencies. Mesh names, authoritative anatomical labels, FMA/UBERON identifiers, and bounds are in `manifest.json`.

Coordinates are in metres. +Y is superior and +Z is anterior. Bounding box is approximately [0.0192893241, 0.7527118, 0.006768436] to [0.04374538, 0.777123868, 0.0325167924]. The source is offset in body-reference coordinates; center the model before scaling for a close-up viewer.

The file contains retina, fovea, macula, optic disc, upper/lower palpebral conjunctiva, vitreous humor, cornea, bulbar conjunctiva, iris, trabecular meshwork, ciliary body, ciliary muscle, aqueous humor, Schlemm's canal, lens, corneoscleral junction, ora serrata, suspensory ligament, ciliary processes, pupil, choroid, and sclera. It does **not** include a separate optic nerve or extraocular muscle mesh. The pupil mesh is an illustrative representation of an opening, not a tissue.

The retina node contains fovea, macula, and optic disc child nodes; the ciliary body contains the ciliary muscle node. Viewers should account for this hierarchy when hiding or isolating a structure.

## Mobile derivative

`human-eye-mobile.glb` is an EvryNest adaptation of the same CC BY 4.0 source. It preserves all 23 anatomical mesh identities and ontology metadata, with 81,851 triangles and approximately 1.32 MB (95% smaller than the original). Changes: removed split vertex normals to weld coincident positions, simplified with glTF Transform 4.3.0 / meshoptimizer at a 0.2 target ratio and 0.002 geometric-error threshold, regenerated smooth area-weighted normals, and quantized position/normal attributes to 14/10 bits. Tiny structures retain more triangles when the error threshold prevents further reduction. This is a visualization derivative, not an unchanged clinical reference.

The derivative uses `KHR_mesh_quantization`, supported natively by Three.js GLTFLoader, and requires no downloadable decoder. Node transforms restore quantized coordinates: viewers must preserve/apply world matrices before flattening meshes. Retina and ciliary-body wrapper nodes were renamed with `_group` to preserve exact selectable mesh names. Verified by loading with the application's GLTFLoader: all 23 mesh names match the original. World-space bounds and checksum are recorded in `mobile-manifest.json`.

## Interactive rendering adaptation

The application adds illustrative procedural surface detail to the licensed geometry: iris radial fibers, collarette and limbal shading, subtle scleral vessel patterns, and a clear glossy corneal material. These generated patterns are not captured biological textures or patient-specific vascular anatomy. The source iris/sclera lack UV coordinates; the renderer maps detail using the model's original anatomical coordinates. Mesh shapes and selectable structures are preserved. Lighting reflections are generated locally from light cards; no external HDR image is loaded. The visualization does not simulate optical refraction or microscopic anatomy.