Text this: Performance of basic networks trained on shape-only images.