Add a ValidBboxesDataset class

[sfmig]

Description

What is this PR

• Bug fix
• Addition of a new feature
• Other

Why is this PR needed?
We want to be able to load bounding boxes tracks (see #167).

What does this PR do??
This PR defines a ValidBboxesDataset class, that follows ValidPosesDataset.

A bounding boxes dataset is made up of 4 data arrays:

• position: (required) holds the x, y coordinates of the bounding box centroid.
  • Typically it will be an array of shape (n_frames, n_individuals, n_spatial_coords), but we only enforce the last dimension (spatial) to have 2 coordinates.
• shape: (required) holds the width and height of the bounding box.
  • The same validation from the position array is applied.
• confidence: (optional) holds the confidence score for the bounding box. If None is passed, it is filled in with nans.
  • We check its n dimensions have the same number of coordinates as the first n dimensions of the position array.
• individual_names: (optional) a list of bounding boxes IDs. If None is passed, we assign 1-based IDs that are unique per frame. The 1-based convention for the IDs follows MOT challenge.

We are assuming 2D bounding boxes only for now (through the validation of the position and the shape arrays).

Questions

• In the tests, I am ensuring the error case is the one I expect to fall in by checking the error message. Should we do this in the rest of the tests too?
• In the pose dataset, we enforce the position array to have a fixed number of dimensions, which I am not doing in the bboxes dataset. This is inconsistent atm, but my understanding from chats with @niksirbi, from this PR Accessor compute_ to only validate required position data variable #206 and from this related issue is that we would like to remove that restriction. Do correct me if I am wrong!

References

Issue #167.

This work is derived from splitting #186 into smaller PRs.

How has this PR been tested?

Tests pass locally and on CI.

Is this a breaking change?

No.

Does this PR require an update to the documentation?

It will require an update once the bounding boxes functionality is complete.

Checklist:

• The code has been tested locally
• Tests have been added to cover all new functionality
• [ n/a ] The documentation has been updated to reflect any changes
• The code has been formatted with pre-commit

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.70%. Comparing base (e4eea35) to head (2f4cf20).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #201      +/-   ##
==========================================
+ Coverage   99.68%   99.70%   +0.01%     
==========================================
  Files          12       12              
  Lines         643      678      +35     
==========================================
+ Hits          641      676      +35     
  Misses          2        2              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[niksirbi]

I like this implementation @sfmig !
I left some nitpicky comments which should be easy to fix.
Regarding your questions, in the order they were posed:

• yes, that is a good idea.
• I think you made the right call here, but this touches on a fundamental issue we three (@sfmig @lochhh) should sit down and discuss when you are back. I attempted to summarise this fundamental issue here: Need to re-design dataset validators? #210

[niksirbi]

Thanks for implementing all the suggestion @sfmig, I only have very few remaining comments, and after that it's good to go.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
No data about Coverage
0.0% Duplication on New Code

See analysis details on SonarCloud