The current tests for the real space projection are insufficient for testing. Among other issues, the object is symmetric, so if we're getting the rotations wrong it's easy to miss.
A few ideas for benchmarks for comparisons:
- off-center object.
- off-center Gaussian: the analytic projection is easy to compute in the real and Fourier domain (it also makes us define some normalization conventions).
- standard outputs of external packages (e.g. RELION). Tests rotation and shift conventions etc. This would also force us to check our axis organization conventions.
The current tests for the real space projection are insufficient for testing. Among other issues, the object is symmetric, so if we're getting the rotations wrong it's easy to miss.
A few ideas for benchmarks for comparisons: