We are experimenting with learning of low-level spatiotemporal classes such as these synthetically generated 16x16-pixel 8-frame snippets.