Of course, they didn't go on to generate "deceptive walking" in the wild. They generate video or pictures as deceptive or non-deceptive and they are able to dis

Of course, they didn't go on to generate "deceptive walking" in the wild. They generate video or pictures as deceptive or non-deceptive and they are able to distinguish the generated stuff without labels once they'd trained a network with labeled data. This might (or might not) serve to prove that these deceptive gestures do exist. The problem of even this is are that there may be a zillion cue characteristic of the subjects being given their directions. But even if such gestures exist, it seems pretty given that this would be overwhelmed the complexity of images captured in the wild, where many other variable govern behavior.