Researchers have developed BenthicDINO, a new framework for processing side-scan sonar images that automatically separates the underlying seabed reflectivity from transient viewing geometries. The framework uses physically motivated augmentations and a penalty term to enforce view-invariance and decouple learned features from physical parameters. In experiments on the S3Seg dataset, BenthicDINO achieved state-of-the-art results with exceptional data efficiency, requiring only 10% of annotated data to reach a mean Intersection over Union (mIoU) of 71.4% and an overall accuracy of 86.5%. The authors attribute these improvements to their dense, hierarchical feature fusion strategy. --- Why it matters: This matters because it enables more efficient processing of side-scan sonar images, which are crucial for mapping complex benthic topographies in marine environments. By reducing the need for manual annotations and improving data efficiency, BenthicDINO has significant implications for researchers working on underwater exploration and monitoring. Source: https://arxiv.org/abs/2608.23215