CNNs trained on places and animacy explain different patterns of variance for the same dataset.