add implementation for label-studio annotations and DeepLabCut

[quantumdot]

Description

Add support for label-studio and DeepLabCut

Data Model Changes

• Fix equality of Points. X and Y coordinates were previously compared using normal float equality semantics. Here we change to utilize the numpy.isclose() method for comparison, where two floats are compared for equality within some tolerance (we use just the default tolerances). Additionally, we will consider that two NaN values are equal, which is normally not true.
• Allow Instance objects to be compared for equality. The points collection within an instance are compared with the rule that both collections of points must contain the same set of nodes, and for each node the point must be equal according to the semantics described above.

Label Studio Support

• support for reading label-studio annotations from a file or list[dict] and conversion to Labels instance via sleap_io.io.labelstudio.read_labels() and sleap_io.io.labelstudio.parse_tasks()
• support conversion of Labels to list[dict] via sleap_io.io.labelstudio.write_labels()
• known issues:
  • label-studio uses relative coordinates (in the range 0-1), so writing label-studio tasks from a labels instance requires that the Video associated with a LabeledFrame properly implement the Video.shape property. If shape is not properly implemented, the label-studio tasks may not function properly.

DeepLabCut Support (single- and multi-animal projects)

• support for reading DLC annotations and conversion to Labels instance via sleap_io.io.dlc.load_dlc() and sleap_io.io.dlc.dlc_to_labels().
• support conversion of Labels to DLC annotations via sleap_io.io.dlc.write_dlc() and sleap_io.io.dlc.labels_to_dlc()
• support determining if the DLC project is a single-animal or a multi-animal style project via sleap_io.io.dlc.is_multianimal()
• support loading Skeleton data from a DLC project via sleap_io.io.dlc.load_skeletons()
• known issues:
  • DLC does not include edges in their skeleton definitions, so a skeleton with only nodes and no edges is constructed.
  • Video data may not be properly implemented.
    • I think there is no way for us to know a video or image shape without us first loading it.
    • We currently only set the frame index for an annotation to zero. probably we can parse it from the filename?

Types of changes

• Bugfix
• New feature
• Refactor / Code style update (no logical changes)
• Build / CI changes
• Documentation Update
• Other (explain)

Does this address any currently open issues?

[list open issues here]

Outside contributors checklist

• Review the guidelines for contributing to this repository
• Add tests that prove your fix is effective or that your feature works
• Add necessary documentation (if appropriate)

Thank you for contributing to SLEAP-IO!

❤️

[codecov]

Codecov Report

Merging #20 (5bd7e4f) into main (d8b7b74) will decrease coverage by 6.78%.
The diff coverage is 78.57%.

@@             Coverage Diff             @@
##              main      #20      +/-   ##
===========================================
- Coverage   100.00%   93.21%   -6.79%     
===========================================
  Files           10       11       +1     
  Lines          526      752     +226     
===========================================
+ Hits           526      701     +175     
- Misses           0       51      +51     

Impacted Files	Coverage Δ
sleap_io/io/dlc.py	69.92% <69.92%> (ø)
sleap_io/model/instance.py	96.90% <78.57%> (-3.10%)	⬇️
sleap_io/io/labelstudio.py	91.20% <91.20%> (ø)
sleap_io/__init__.py

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[talmo]

Looking great @quantumdot! Feel free to mark this ready for review for @roomrys whenever you're done.

[quantumdot]

@talmo @roomrys I believe I have the bulk of functionality implemented for DLC and label-studio. Perhaps it is a good time to review changes?

We should also discuss Video instances, as some outstanding issues depend on the implementation details there.

[talmo]

@talmo @roomrys I believe I have the bulk of functionality implemented for DLC and label-studio. Perhaps it is a good time to review changes?

We should also discuss Video instances, as some outstanding issues depend on the implementation details there.

Thanks! @roomrys can do the first review. Let's work around the Video stuff for now since that depends on the very unfinished branch at https://github.com/talmolab/sleap-io/tree/talmo/pims-video-backend.

[quantumdot]

CI / Tests are failing on python==3.7 it seems due to pickle protocol version mismatch. See related: DeepLabCut/DLCutils#19. Do you have a DLC dataset which works across the python version 3.8 boundary, where pickle version changed from 4 -> 5?

CI / Tests fail on windows python==3.9 due to an import error for pytables. I don't understand why.

[talmo]

Hi @quantumdot,

Ok, maybe let's split this up into two PRs? You can copy over dlc.py and the test data into a new branch where we can start to work around some of these version issues.

Talmo

[quantumdot]

Closing this PR as it is superseded by the following (these changes split into three separate PRs based on separation of concerns)

• Changes to Point and Instance equality semantics: Change equality semantics for Points and Instances #21
• Add support for DeepLabCut projects: Add DeepLabCut project support #25
• Add support for LabelStudio projects: Add LabelStudio project support #26