Two new papers on video classification, and why the network that only sees motion beat the one that sees the frames.