Vert-NSM: neural shape models for shape completion, classification, and landmark-free morphometrics of vertebrate specimens
Abstract
Natural history museums are building digital 3D archives at a scale that enables geometric morphometric analysis across thousands of specimens. Landmark-based analyses depend on anatomical priors and often exclude unidentified and incomplete specimens, which are abundant in collections and the fossil record. Continuous shape representations offer a landmark-free alternative that can be combined with neural networks to perform shape completion and classification. It is unknown how their morphospaces compare with landmarks or how accurately they perform. We present Vert-NSM, neural shape models for vertebrate biodiversity specimens, and apply them to squamate vertebrae, a well-represented, but challenging test case. Models were trained on 2,896 vertebral meshes from 95 specimens across 40 families, produced in part by the oVert digitization initiative. We evaluated shape completion, classification, and morphometrics of sparse landmarks, dense correspondences, and Vert-NSM latents. Shape completion accuracy reached 0.375% of a vertebral radius. Classification exceeded frequency-weighted chance at nearly every level tested, reaching 80% accuracy for the cervical spine and 90% for Cordylidae. Vert-NSM shared the dominant axes of variation recovered by landmarks, and identified shape differences that did not depend on anatomical priors. In downsampling experiments, models performed well when trained with at least 30 specimens. Vert-NSM has the potential to unlock discoveries from vertebrate collections at scale and in deep time.
Related
- Towards a unified biology of vertebrate number sense
- Stochastic Completeness for Landmark Shape Spaces with Induced Metrics
- Topologically Relevant Landmark Selection on Point Clouds
- A General Kernel Framework for Non-CND Distance Measures Using |D|-Dimensional Sparse Landmark Embeddings
- Dynamic risk assessment by landmark modeling of the restricted residual mean survival time.
BibTeX
@misc{wolcott2026vertnsm,
title = {{Vert-NSM: neural shape models for shape completion, classification, and landmark-free morphometrics of vertebrate specimens}},
author = {Katherine Aubry Wolcott and Amar Cutura and Matthew Delage and Kelsey Fenner and Ali Sean Khan and Jia Xuan Koh and Mitchell Riegler and Colin Sawyer and Joshua Yu and Anthony Gatti and Edward Stanley and Arthur Porto},
howpublished = {EcoEvoRxiv},
year = {2026},
month = oct,
doi = {10.32942/x2nt3x},
url = {https://ecoevorxiv.org/repository/view/15649/},
note = {CC BY-NC-SA 4.0},
}
RIS
TY - UNPB
TI - Vert-NSM: neural shape models for shape completion, classification, and landmark-free morphometrics of vertebrate specimens
AU - Katherine Aubry Wolcott
AU - Amar Cutura
AU - Matthew Delage
AU - Kelsey Fenner
AU - Ali Sean Khan
AU - Jia Xuan Koh
AU - Mitchell Riegler
AU - Colin Sawyer
AU - Joshua Yu
AU - Anthony Gatti
AU - Edward Stanley
AU - Arthur Porto
PY - 2026
DA - 2026/10/22
PB - EcoEvoRxiv
DO - 10.32942/x2nt3x
UR - https://ecoevorxiv.org/repository/view/15649/
AB - Natural history museums are building digital 3D archives at a scale that enables geometric morphometric analysis across thousands of specimens. Landmark-based analyses depend on anatomical priors and often exclude unidentified and incomplete specimens, which are abundant in collections and the fossil record. Continuous shape representations offer a landmark-free alternative that can be combined with neural networks to perform shape completion and classification. It is unknown how their morphospaces compare with landmarks or how accurately they perform. We present Vert-NSM, neural shape models for vertebrate biodiversity specimens, and apply them to squamate vertebrae, a well-represented, but challenging test case. Models were trained on 2,896 vertebral meshes from 95 specimens across 40 families, produced in part by the oVert digitization initiative. We evaluated shape completion, classification, and morphometrics of sparse landmarks, dense correspondences, and Vert-NSM latents. Shape completion accuracy reached 0.375% of a vertebral radius. Classification exceeded frequency-weighted chance at nearly every level tested, reaching 80% accuracy for the cervical spine and 90% for Cordylidae. Vert-NSM shared the dominant axes of variation recovered by landmarks, and identified shape differences that did not depend on anatomical priors. In downsampling experiments, models performed well when trained with at least 30 specimens. Vert-NSM has the potential to unlock discoveries from vertebrate collections at scale and in deep time.
ER -