New idea: Guided 2D. Doesn’t entirely supercede the need for templated 2D (which can be “forced” to a degree with a large population of particles of one class added to a small pool and removed later in particle sets), but has less bias and several use cases. for both streaming and standalone. after some early (or tunable) number of iterations, user suggests which 2D are more correct than others with green=good red=bad unselected=undefined, and somehow cryosparc learns that these are the preferred representations of the particles. Can be repeated after 5, 10, 15 iterations. Use cases: 1) keep 2Ds from slipping out of center especially for hollow spheres 2) use it iteratively to “recenter” particles like half a proteasome or a subdomain not center-of-mass 3) select for rare views which get lost in large particle sets due to the ~200 class limitation 4) etc.