Vert-NSM Morphospaces Reference anatomy Shape variation Data & code

Interactive figures

Morphospaces, reference anatomy and shape-variation videos accompanying Vert-NSM.

Vert-NSM: neural shape models for shape completion, classification, and landmark-free morphometrics of vertebrate specimens.
Wolcott et al. 2026

Code Models & data Micro-CT

One vertebra, decoded along PC1 of the latent space

Explore the morphospaces

Every ordination in Figure 6, live. Hover a point for its species and vertebra; drag to zoom. All 2,316 training vertebrae are plotted.

Loading the plot…
Open in a new tab

Landmark representations concentrate variance in early components (PC1–PC2: 53.6% sparse, 66.2% dense) where NSM latents spread it (20.2%), so the latent space looks unstructured under PCA and resolves under t-SNE and UMAP. Amphisbaenia separates in every panel.

The reference anatomy, in 3D

The mean vertebra, with the landmark schemes Vert-NSM is compared against. Drag to rotate, scroll to zoom.

Loading the model…
Open in a new tab

Shape variation, moving

Latents displaced along each principal component from the specimen nearest the mean, then decoded — four views at once.

PC1 — elongation

The dominant axis in every representation: position along the spinal column.

PC2 — tall to wide

The neural spine reduces as the centrum broadens and the postzygapophyses thicken.

PC3 — leaving the manifold

Geometry turns implausible at the extremes, where training data is sparse.

PC4

Recovers the variation that landmark PC3 describes (|ρ| = 0.70).

PC5

Beyond the components reported in the paper.

The whole dataset

All 2,896 meshes in sequence, labelled. Long — press play.

Data & code

Everything behind these figures is open.