angular_project_library with non-trivial psi_sampling failing

[jvdoorss]

I found that, when using angular_project_library, the image output of the projections is not as expected and in particular not in sync with the corresponding angular metadata. I suspect it being due to the indexing in this line:

xmipp/src/xmipp/libraries/reconstruction/angular_project_library.cpp

Line 250 in 84a87d4

P.write(output_file,(size_t) (numberStepsPsi * i + mypsi +1),true,WRITE_REPLACE);

To reproduce:

I did

xmipp_angular_project_library -i volume.vol -o images.mrcs --sampling_rate 22.5 --psi_sampling 22.5

on a 100x100x100 volume. 22.5 degrees was chosen for easy testing as kind of the the largest divisor of 360 below the sampling cut-off.

It produces an images.doc file with $92 \times 16 = 1472$ where 92 is the number of spatial directions (also the size without psi_sampling) and 16 = 360 / 22.5 the number of in-plane rotations. So far so good, this is exactly what you would want/expect. ✅

images.mrcs however contains 1703 images ($\neq 1472$) of whom all except 226 are empty. A possible explanation, consistent with the indexing in the line quoted above, is the following:

The (0-based) indices of this non-empty set correspond exactly to the unique values in the set
$\lbrace floor(i)+15 * j|i\in \lbrace 0,22.5,45,...337.5\rbrace, j\in \lbrace 0,...91\rbrace\rbrace$
with i, j indexing in-plane and spatial rotations, respectively and 15 being an (incorrect) offset numberStepsPsi. While the double loop spans the correct 1472 cases, the indices will go from 0 to 1702 = 91 * 15 + 337 but contains only 226 unique indices.

Possible fix, I think:
Using a counter rather than the angle mypsi clearly was the intention. Doing that correctly and ensuring numberStepsPsi is right, would yield the correct result here.

However, I guess adding an intrinsic label to the Sampling, used both when writing the images and angular metadata, would allow to avoid any mismatch and easily check unicity and give a more robust implementation? 🤔

[jvdoorss]

Additionally, I think that replacing the 16 values of mypsi with the integers [0...15] would store the images in transposed order (psi as the inner counter, changing at every step) compared to the .doc metadata (psi the outer counter, changing every 92 samples).