Exclude projections

About

This MR extends the ignore_projection parameter to make it more user-friendly.

Currently, one has to provide projections indices to ignore from the scan. This is quite cumbersome ; most of the time users want to exclude an angular range or a set of angles.

API

The extended parameter can be:

1. ignore_projections={"indices": indices_list}: ignore projections from their index (previous behavior)
2. ignore_projections={"angles": angles_list}: ignore projections from their angle (in degrees)
3. ignore_projections={"range": angular_range_tuple}: ignore an angular range

When passing ignore_projections=indices_list (previous behavior), we fall-back on case (1) above with a deprecation warning.

Implementation

In TomoScan objects, the projections are primarily identified by their index. For, example self.projections is a dict in the form {index: data_url}. So internally, everything will be done using projections indices.
So when passing a list of angles, one will need a mapping angle <-> proj_index.

In NXTomoScan, the indices are directly accessible with image_key (ex. np.where(image_key == 0)). On the other hand, angles are also accessible in self.rotation_angle. So the mapping will be easy to do.

In EDFTomoScan, things are a bit more complicated:

• get_proj_urls() has to be called first to get both indices and projection URL. We don't know all the projection indices until the full list is built:
  • There might be "jumps" in the file suffix used for index (eg. frame_0010.edf, frame_0012.edf)
  • The list of DataUrl is built using os.listdir() which returns files in arbitrary order
• Therefore, the mapping index <-> angle cannot be done early at initialization.
• Once the list of projections is built, _reload_projections() can be used to insert the "ignore projection" parameter