Researchers found that infant pose was accurately predicted in cross-validation, held-out frames, and held-out infants with respective receiver operating characteristic area under the curves.