# How reliable is an embedding of the seabed? Repeat-look measurement of frozen encoders on synthetic aperture sonar

DevFeed: [How reliable is an embedding of the seabed? Repeat-look measurement of frozen encoders on synthetic aperture sonar](<https://devfeed.tech/articles/how-reliable-is-an-embedding-of-the-seabed-repeat-look-measurement-of-frozen-encoders-on-synthetic-aperture-sonar-125251.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10770>)

Author: Harold Horsley, Oscar Pizarro, Hayat Rajani, Morten Omholt Alver

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [ai evaluation](<https://devfeed.tech/topics/ai-evaluation.md>), [clustering](<https://devfeed.tech/topics/clustering.md>), [Image](<https://devfeed.tech/topics/image.md>)

Tags: [acquisition](<https://devfeed.tech/tags/acquisition.md>), [classification](<https://devfeed.tech/tags/classification.md>), [clustering](<https://devfeed.tech/tags/clustering.md>), [clusters](<https://devfeed.tech/tags/clusters.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [coral](<https://devfeed.tech/tags/coral.md>), [correlation](<https://devfeed.tech/tags/correlation.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [habitat](<https://devfeed.tech/tags/habitat.md>), [identity](<https://devfeed.tech/tags/identity.md>), [image](<https://devfeed.tech/tags/image.md>), [linear](<https://devfeed.tech/tags/linear.md>), [look](<https://devfeed.tech/tags/look.md>), [measurement](<https://devfeed.tech/tags/measurement.md>), [measurements](<https://devfeed.tech/tags/measurements.md>), [repeat](<https://devfeed.tech/tags/repeat.md>), [sonar](<https://devfeed.tech/tags/sonar.md>), [stat-ap](<https://devfeed.tech/tags/stat-ap.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>)

## AI overview

The study finds that standard single-look metrics can miss poor repeat-look reliability in seabed embeddings. On a synthetic-aperture sonar survey, frozen DINOv3 ViT-S/16 had a distance-based intraclass correlation of 0.20, and its nearest neighbours and clusters barely agreed across looks, despite stable standard metrics and a coral probe with AUC 0.93. The probe's learned direction was more reliable (ICC 0.60), suggesting habitat-level classification can work while patch-identity uses such as cross-pass comparison and retrieval fail for this dataset. A sonar-pretrained encoder, BenthicDINO, and a classical SAS texture descriptor showed the same pattern. Generalizability-theory analysis attributed less than a tenth of within-cell spread to look geometry; the remaining cell-specific spread could not be separated without repeat passes at matched geometry. The authors argue repeat-look reliability should accompany seafloor representation evaluations.

## Source excerpt

arXiv:2611.10770v1 Announce Type: cross Abstract: Pretrained encoders from self-supervised representation learning are increasingly applied to sonar imagery of the seafloor, where they are evaluated with protocols that take one embedding per target. A target patch of seabed, by contrast, is re-imaged from an alternate look on overlapping survey passes, so each embedding is one measurement of that target. Standard protocols cannot say how consistent those measurements are, or how much of their spread belongs to the patch and how much to the acquisition geometry. We treat this as a measurement-reliability problem. Using a synthetic-aperture sonar (SAS) survey, we measure the repeat-look reliability of embeddings from a frozen natural-image encoder (DINOv3 ViT-S/16). Standard single-look metrics, effective rank, clustering and a linear probe for coral, are all stable to the look, and the probe performs well (AUC 0.93). Yet the repeat-look reliability of the same embeddings is poor, with a distance-based intraclass correlation (dbICC) of 0.20 and nearest neighbours and clusters that barely agree across looks. This contradiction resolves as we find the probe learns a stable direction, whose ICC of 0.60 is 3.0 times that of the full embedding. Variance components, calculated from generalizability theory, attribute under a tenth of the within-cell spread to look acquisition geometry. The rest is cell-specific and cannot be separated without repeat passes at matched geometry. A sonar-pretrained encoder (BenthicDINO) and a classical SAS texture descriptor show the same pattern. For this dataset and the tested representations, uses that depend on patch identity fail, such as cross-pass comparison and retrieval, while uses at the level of habitat type (e.g. coral classification) do not. Because the standard metrics did not detect this, we argue that repeat-look reliability should accompany any evaluation of representations of the seafloor.