Assessing the determinants of larval fish strike rates using computer vision

•We present an AI-assisted pipeline for measuring rare fitness-determining behaviors.•The pipeline successfully detected feeding strikes of larval fish from untrimmed videos.•The pipeline was ×10 times faster than human annotation and detected twice as many events.•Strike rates increased with temperature, but was not age-dependent.•This is the first direct measurements of larval strike rates outside the laboratory.