Beluga Whales

Data-Driven Whale Counts: AI Advances Arctic Beluga Monitoring

Manual annotation of aerial drone imagery remains one of the steepest bottlenecks in Arctic beluga monitoring, and calf detection only compounds the challenge. This study offers a practical way forward. By testing semi-supervised detection models against supervised baselines across annotation budgets, the authors show that SEMI-DETR achieves comparable results with 5- to 10-fold less labeled data. The 24.3% relative improvement in calf detection at 1% labeling is particularly encouraging, as it directly supports reproductive health assessments.