Some ideas * start from simple object: * e.g. sphere: No matter how align, the projections will give the right alignment, since they are all the same. * have more complicated object, e.g. several constant boxes of ones of length 10-20 in a total size of n_pix=128 * corrupt initial volume to start with a noisy estimate (e.g. low pass filter by using 3d fft with filter, or add in white noise in 3d) * generate slices (e.g. 1000) * call iterative_refinement. Not sure what to pick for sigma, so might want to keep an eye on that. * check for the following on the returns * assert that the half maps are close within some threshold: (half_map_1 * half_map_2).sum()**2 / (half_map_1**2.sum()) * (half_map_2**2).sum()) > 0.99
Some ideas