Sampling video into sequences of still imagery to analyze using deep learning can save storage and computational costs and creates new opportunities for non-consumptive analysis, but raises the question of what the ideal sampling rate is for each kind of analysis.