WildFin: An In-the-Wild Video Dataset for Fish Behavioral Recognition

Abstract

Recent advances in field technology have led to a massive influx of in-the-wild video data for ecological science. The primary bottleneck in leveraging this data is the high cost of expert annotation. While computer vision offers a potential solution, current models frequently fail when deployed in complex marine environments.

WildFin characterizes these failures through a dataset collected and annotated by ecologists. It includes two critical deployment paradigms: stationary cameras monitoring groups of fish and dynamic divers following individual subjects. The benchmark evaluates modern vision foundation models and quantifies the remaining gap between current model capabilities and real-world underwater behavioral analysis.