BenthicDINO: Physics-Informed Self-Distillation for View-Invariant Side-Scan Sonar Representations

Name: BenthicDINO: Physics-Informed Self-Distillation for View-Invariant Side-Scan Sonar Representations
Uploaded: 2026-10-20T17:47:12.843394+00:00
Duration: 238 s
Description: BenthicDINO creates view-invariant representations for side-scan sonar imagery by using a physics-informed self-supervised learning framework. It enforces physical constraints and statistical independence between features and viewing parameters, allowing the model to learn robust seabed features with only 10% of labeled data, achieving near peak performance.

BenthicDINO creates view-invariant representations for side-scan sonar imagery by using a physics-informed self-supervised learning framework. It enforces physical constraints and statistical independence between features and viewing parameters, allowing the model to learn robust seabed features with only 10% of labeled data, achieving near peak performance.

Open the full episode