CoralLite: μ \mu CT Reconstruction of Coral Colonies from Individual Corallites
Jess Jones, Leonardo Bertini, Kenneth Johnson, Erica Hendy, Tilo Burghardt · 2026
Read CoralLite: μ \mu CT Reconstruction of Coral Colonies from Individual Corallites by Jess Jones, Leonardo Bertini, Kenneth Johnson, Erica Hendy, Tilo Burghardt (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-15093.txt.
Introduction
Jess Jones and four others present CoralLite, a micro-CT dataset rebuilding whole coral colonies from single corallites, the skeletal cups around each polyp. Colony history sits in growing skeletons that may span centuries, while living tissue is only a thin outer layer. CoralLite pairs fully measured 3D segment maps with cross-slice linking into 3D models per corallite. A hybrid V-Trans-UNet model, pre-trained weakly and fine-tuned on over 8 thousand manual labels, reaches 0.94 topological accuracy on unseen slices.
Catalog record
| CREATOR | Jess Jones, Leonardo Bertini, Kenneth Johnson, Erica Hendy, Tilo Burghardt |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34499 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.15093
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
CoralLite: μ \mu CT Reconstruction of Coral Colonies from Individual Corallites Jess Jones, Leonardo Bertini, Kenneth Johnson, Erica Hendy, Tilo Burghardt Abstract The life history of an individual coral is archived within the accreting skeleton of the colony. While reef-forming coral colonies (e.g. massive Porites sp.) may live for hundreds of years and deposit calcareous structures many metres in height and width, their living tissue is a thin outer surface layer comprised of asexually-dividing polyps that only survive a few years. To understand the rate and timing of polyp division and the consequences for colony skeletal growth, scientists need to track the skeletal corallite deposited around each polyp. Here we propose CoralLite, an annotated μ \mu CT scan dataset of entire calcareous skeletons and an associated, first corallite deep learning reconstruction baseline. CoralLite combines fully quantified volumetric segmentations with cross-slice linking for visualisations of 3D models for each corallite up to colony scale. For segmentation, we propose and evaluate in detail a hybrid V-Trans-UNet architecture applicable to segmenting tiled μ \mu CT virtual slabs of Porites sp. colonies. The model is pre-trained on weakly annotated data and topology-aware fine-tuned using fully annotated slice sections with 8k+ manual corallite region annotations. On unseen slices of the same colony, the resulting model reaches 0.94 topological accuracy at mean Dice scores of 0.77 on the same colony and projection axis, and 0.63 mean Dice scores on a different, biologically unrelated specimen. Whilst our experiments are limited in scale and context, our results show for the first time that visual machine learning can effectively support full 3D individual corallite modelling from μ \mu CT scans of coral skeletons alone. For reproducibility and as a baseline for future research we publish our full dataset of 697 μ \mu CT slices, 37 partial or full slice annotations, and all network weights and source code with this paper. 1 Introduction Scientific Motivation. Scleractinia , or stony corals, can form colonies by asexual division of individual polyps. Each polyp deposits a surrounding protective tube of aragonite ( CaCO_{3} ), known as a corallite, which forms a porous wall between neighbouring polyps and collectively creates the lasting skeletal structure of the colony [ 7 , 9 ] . For example, an individual massive Porites sp. colony can grow large enough to form a micro-atoll, or alternatively a dome-shaped skeletal structure that can be over 5 metres in diameter, and encompass over 500 years of growth [ 11 , 14 ] . Fossils of Scleractinia date back to the middle Triassic period and have therefore been a feature of ocean ecosystems for more than 230 million years. Since the Miocene, the genus Porites has been functionally important in shallow, tropical and sub-tropical waters as one of the key reef-framework builders of warm-water coral reefs [ 2 ] . Massive Porites sp. are also the most commonly studied coral for palaeoclimate reconstructions and growth rate variability associated with environmental change [ 17 ] . In \mu CT scans, the structure of these colonies appears as large stacks of cross-sectional slices containing many small interlocked corallites as shown in Figure 1 . Manual tracing of the internal space within individual corallites through these volumes is possible and practised today for fine-grained skeletal analysis, however this approach is time-consuming and laborious [ 15 ] . Porites sp. serve as an excellent case study for analysis automation with direct applicability to vital coral reef research. Figure 1: Overview of CoralLite. The proposed pipeline comprises three main stages: (1) Novel Dataset: Sequential 2D cross-section slices from 3D \mu CT scans of a Porites sp. coral colony are annotated and processed into 5x224x224 tiled snippets. (2) Volumetric Segmentation: A 3D-context-aware Trans-UNet pipeline with a hybrid ResNet50-ViT backbone (details in Fig. 4 ) generates corallite segmentations using a topological loss for accurate boundary prediction. (3) Corallite Reconstruction: Segmented tiles are stacked into a volume to generate a 3D corallite-level colony model. Research Aims and Contributions. This paper considers whether such identity-preserving reconstructions of individual corallites in 3D from \mu CT data alone can be automated using deep learning for animal biometrics [ 13 ] . The approach is outlined in Figure 1 . In contrast to most existing coral reconstruction research, we attempt fully automated segmentation of individual corallites of a scanned colony and quantify baselines of this task here for the first time. Our aim is not to present a complete biological model of coral growth at this stage since expert annotations are limited and expensive to produce at scale. Instead, our aim is to assess whether a reasonably sized, annotated dataset and volumetric learning together are sufficient to support an automated colony reconstruction at corallite level which is useful to practitioners today and sets baselines for the community to build on this work. Recent progress in medical and volumetric image segmentation in particular presented a useful starting point for our work [ 4 , 10 ] . Procedurally, we trained 3D-aware segmentation models to identify corallite regions encoded in binary segmentation maps, and then apply cross-slice linking procedures supported by topological learning constraints to assemble predicted regions into per-corallite, identity-preserving 3D models. In line with UNet architecture tests for corallites in [ 5 ] , we build on hybrid Video-TransUNet architectures and research [ 3 , 22 ] that combines local texture modelling with broader contextual reasoning. Our ResNet–ViT baselines follow this paradigm and operate on stacks of adjacent coral \mu CT slices such that cross-slice aggregation can exploit continuity in 3D data. Our contributions are: • Annotated Dataset: 697 \mu CT scan slices plus a set of 8,412 manual corallite ground truth segmentations in full slices of massive Porites complemented further by point-annotation corallite priors across 33 of the slices; • Topology-guided 3D Reconstruction Pipeline: a volumetric segmentation pipeline based on a V-TransUNet model with cross-slice spatial feature context learning, together with a topological loss for discouraging merged regions with full release of key code and network weights; • Baseline Results and Evaluation: baselines with a first task evaluation across same-colony, cross-axis, and cross-specimen test settings; and • Visualisation Bridge: a simple cross-slice tracing and geometry generation procedure for reconstructing individual corallites as 3D objects suitable for manual inspection by marine scientists using standard graphics packages; 2 Related Work 3D Coral Reconstruction in Marine Biology. Recent research has shifted toward volumetric 3D imaging techniques, such as computed axial tomography (CT) and micro-CT ( \mu CT), to precisely measure coral extension rates and corallite demography [ 21 , 15 ] . These high-resolution reconstructions allow for a more accurate, non-destructive assessment of individual polyp longevity and skeletal growth trajectories compared to traditional 2D analysis [ 1 , 7 ] . By resolving the 3D spatial organisation of coral modules, these methods hold rich, previously inaccessible information about colonial developmental dynamics [ 15 ] . Corallite Reconstruction Automation. Automated computer-vision pipe-lines for individual corallite reconstruction from whole-colony \mu CT remain scarce: most coral \mu CT work still emphasises manual processing, rather than learned, 3D corallite models. A notable recent exception [ 5 ] applies deep semantic segmentation (U-Net family models, including Attention U-Net) to \mu CT of scleractinians such as Montastrea cavernosa and Porites astreoides ,