v0.14.0

To update movement to the latest version, see the update guide.

What's Changed

⚡️ Highlight: unified data loading interface

• Unify loaders by @lochhh in #722

Thanks to @lochhh's tireless efforts, loading poses and bounding boxes is now handled by a single entry point, movement.io.load_dataset.

The new load_dataset function works for all our supported third-party formats. For example:

from movement.io import load_dataset

# DeepLabCut -> poses dataset
ds = load_dataset("path/to/file.h5", source_software="DeepLabCut", fps=30)

# SLEAP -> poses dataset
ds = load_dataset("path/to/file.slp", source_software="SLEAP")

# NWB -> poses dataset
ds = load_dataset("path/to/file.nwb", source_software="NWB")

# VGG Image Annotator tracks -> bounding boxes dataset
ds = load_dataset("path/to/file.csv", source_software="VIA-tracks")

Similarly, movement.io.load_multiview_dataset replaces the old movement.io.load_poses.from_multiview_files, with added support for bounding boxes:

from movement.io import load_multiview_dataset

# Load LightningPose pose predictions from two cameras.
ds = load_multiview_dataset(
    {"cam1": "path/to/cam1.csv", "cam2": "path/to/cam2.csv"},
    source_software="LightningPose",
    fps=30,
)

Software-specific loaders (e.g. load_poses.from_dlc_file, load_bboxes.from_via_tracks_file) remain available for users who want full control over the loading process.

Warning

The following functions are deprecated and will be removed in a future release:

• load_poses.from_file → use load_dataset instead
• load_bboxes.from_file → use load_dataset instead
• load_poses.from_multiview_files → use load_multiview_dataset instead

Migrating from deprecated functions:

# Before
from movement.io import load_poses, load_bboxes

ds = load_poses.from_file("file.h5", source_software="DeepLabCut", fps=30)
ds = load_bboxes.from_file("file.csv", source_software="VIA-tracks", fps=30)
ds = load_poses.from_multiview_files(
    {"cam1": "cam1.csv", "cam2": "cam2.csv"},
    source_software="LightningPose", fps=30,
)

# After
from movement.io import load_dataset, load_multiview_dataset

ds = load_dataset("file.h5", source_software="DeepLabCut", fps=30)
ds = load_dataset("file.csv", source_software="VIA-tracks", fps=30)
ds = load_multiview_dataset(
    {"cam1": "cam1.csv", "cam2": "cam2.csv"},
    source_software="LightningPose", fps=30,
)

🐛 Bug fixes

• Adapt savgol_filter for compatibility with Scipy >= 1.17 by @niksirbi in #761
• Fix coordinate assignment for elem2 in _cdist by @HARSHDIPSAHA in #776
• Hide _factorized properties from Points layer tooltips by @niksirbi in #781

🤝 Improving the contributor experience

• Docs: restructure contributing guide and add community cards by @ishan372or in #764
• Add preliminary benchmarks by @sfmig in #772
• Use tox-uv in test action by @niksirbi in #766
• Move docs dependencies to pyproject.toml by @AlgoFoe in #774

📚 Documentation

• Fix docstring Raises formatting in LineOfInterest.normal by @AlgoFoe in #771
• Added link to FOSDEM 2026 talk by @niksirbi in #797
• Fix broken 404 page by @AlgoFoe in #785
• Fix links to SLEAP analysis h5 docs by @niksirbi in #807
• Add code snippets for verifying sample data loading by @Edu92337 in #759

🧹 Housekeeping

• Restrict sphinx version to < 9 by @niksirbi in #768
• Unpin sphinx and pin ablog>=0.11.13 by @niksirbi in #811
• Fix deprecated license syntax by @sfmig in #549
• Ignore Docutils URL in linkcheck by @lochhh in #793
• Ignore ISO URL in linkcheck by @lochhh in #815
• Authenticate with GitHub token during linkcheck by @niksirbi in #800
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #795
• Contributors-Readme-Action: Update contributors list by @github-actions[bot] in #787
• Bump conda-incubator/setup-miniconda from 3.2.0 to 3.3.0 by @dependabot[bot] in #788

New Contributors

• @AlgoFoe made their first contribution in #771
• @HARSHDIPSAHA made their first contribution in #776
• @ishan372or made their first contribution in #764
• @Edu92337 made their first contribution in #759

Full Changelog: v0.13.0...v0.14.0