I don't think it does kill your point. The set of images used in the sample are obviously biased. They are by people posting pictures they hope to use to attrac

I don't think it does kill your point. The set of images used in the sample are obviously biased. They are by people posting pictures they hope to use to attract others of a certain gender (or genders). Even if it is specifically taking dimensional measurements surely some facial expressions bias the dimensions. This could still be the equivalent of a duck-face detector.