star_catalog_IDs issue - Githubissues

esa / tetra3

A fast lost-in-space plate solver for star trackers.

https://tetra3.readthedocs.io/en/latest/

Apache License 2.0

102 stars 24 forks source link

star_catalog_IDs issue #9

Closed astropatel closed 1 year ago

astropatel commented 1 year ago

Dear Gustav,

I'm trying to generate my own database using a fake star catalog. I pulled a version of tetra3 about a month ago, and I was able to generate a database using my list of stars (though finding a solution has been NaNs thus far). In any case, I pulled the latest version with the new star_catalog_IDs property and when I try to load the new database into a new instatition of tetra3, I run into a ValueError concerning pickle files. I've changed the code a bit to include my new test catalog and tried to incorporate the star_catalog_IDs as best as possible -- but am wondering if that property is necessary to the overall function of the calculation?

gustavmpettersson commented 1 year ago

Hi,

Could you upload this database somewhere for me to look at? (Or a small one with the same behaviour)

To be clear

The database you are loading, was it built with the latest version or the older one?
To load the database, were you using the latest version or the older one?

It's possible that this is from an old bug where we would store a None in the star_catalog_IDs, making them unable to be unpacked without pickle, but that should not be there anymore.

astropatel commented 1 year ago

I'm uploading the database as well as the version of the code I'm using (with the changes i've made) and a test image..

The database was built and loaded using the latest version. In the new one, it's having issues with the pickle error. In the old and new one, I get a bunch of nan's when trying to use solve_from_image.

rahul_tetra3.zip st version.

gustavmpettersson commented 1 year ago

Without looking in detail, it appears that you are using 'str' type for the star catalogue ID. This cannot be used in a np archive without pickling. You need to make the ID a native np type, e.g. U16 for a 16 byte char array

astropatel commented 1 year ago

That fixed it! Thank you. I am still having a problem with obtaining solutions from my test images.

gustavmpettersson commented 1 year ago

Great.

I don't know what you are trying to do, but at least these things need to be true for it to work:

The database you use to build the catalogue needs to have at least 20 stars per FOV.
The FOV selected when building the database needs to match the FOV of your test images.
The centroids need to be correctly extracted from the test images.
The test images need to be correctly projected (not distorted).