Tags

Tags give the ability to mark specific points in history as being important

  • v4.1.1

    protected
    e059ea96 · Update package version · 
    This version includes:

- new computation (and related visualisation) correlating granularity values to annotation timestamps;
- renaming lenient IoU to IoM everywhere in the code and docs;
- new docstrings.

  • v4.1.0

    protected
    4fdb1c5f · Update package version · 
    This version includes fatigue evaluation computation and the related visualisation.

Breaking changes: more functions which were marked as deprecated were removed.

  • v4.0.0

    protected
    646fda67 · Restructure repo for publication to PyPI · 
    This version includes:

- a new special macro-category name "Excluded" to exclude certain categories from the scale complexity computation;
- more brief functions for reduced output noise;
- extended valid names for macro-categories;
- an implementation of C-Score rewritten from the ground up;
- a new visualisation showing both categories and macro-categories together for better understanding scale complexity results;
- zooming synced through all subplots;
- a new implementation of count and size granularities;
- various quality of life improvements.

Fixes:

- fix alpha_per_image not being computed if De = 0.0 (a8c9a470)
- scale complexity was not being computed for single images (07eb52ed)
- the biggest bounding box was not being counted (ffcef802)

Breaking changes:

- functions marked as deprecated in previous versions have been removed.

  • v3.0.1

    protected
    d097120d · Update package version · 
    Changelog:

- fixed a bug in which same-annotator annotation were checked normally in Alpha calculation;
- added new "brief" function to Result to print essential information only;
- reorganised notebooks and added new notebooks for experimental results.

  • v3.0.0

    protected
    e1186a93 · Update package version · 
    Changelog:

- add new RandomGroundTruth model which allows the user to create random ground truth by sampling real ground truth;
- implemented inheritance throughout the package to make RandomGroundTruth and GroundTruth interoperable;
- move sampling functions to own file in the utils submodule;
- remove unnecessary steps found in the computations of granularity.py and rater_vitality.py
- allow only name-wise GT selection in granularity.py and rater_vitality.py to allow interoperability with RandomGroundTruth;
- fix the initialisation of BoundingBox with the correct type;
- update the default Kappa for zero counts to 0.0;
- smaller optimisations, such as deletion of useless imports.

  • v2.5.0

    protected
    13b2a8fa · Update package version · 
    Changelog:

- add new granularity metrics for both size of bounding boxes and number of bounding boxes per image;
- move bounding box area calculation to a property in the BoundingBox object.

  • v2.4.0

    protected
    44cd922c · Merge branch 'feature/loc_agreement' into 'master' · 
    Changelog:

- introduced two new metrics: L-Score and C-Score, for localisation and categorisation agreement;
- fixed Recognition metric;
- fixed a circular import on Result;
- fixed a rounding error on iou_sweep.

  • v2.3.0

    protected
    037b5f9c · Update package version · 
    Changelog:

- added Rater Vitality metric from McCulloh et al.;
- added function to collapse categories into macro-categories (both interactively and non-interactively);
- update Alpha computation to support macro-categories;
- added Recognition metric from McCulloh et al.;
- added Scale Complexity metric from McCulloh et al.

  • v2.2.7

    protected
    82fbe188 · Update package version · 
    Changelog:

- updated Ground Truth files to fix wrong bounding box format;
- removed bounding box-fixing code from import phase of annotations;
- fix counting images that are empty for more than two annotators;
- fix a piece of text in the stats shown on images.

  • v2.2.6

    protected
    890da39e · Update package version · 
    Changelog:

- update MIN_ALPHA value to 0.0.

  • v2.2.5

    protected
    d3dce337 · Update package version · 
    Changelog:

- added new Lenient IoU metric for IoU computation;
- add new boolean field to Result to show whether Lenient IoU was used;
- change show_iou_sweep title dynamically to show whether Lenient IoU was used;
- add new area calculation function in iou.py;
- add support for Lenient IoU to all metrics.

  • v2.2.4

    protected
    a5e46ba4 · Update package version · 
    Changelog:

- add new function which computes Alpha and Kappa for a range of IoU values;
- add new function which plots Alpha and Kappa values against IoU thresholds on a graph;
- add safeguards to the computation of the global Alpha metric.

  • v2.2.3

    protected
    5bcd1b62 · Update package version · 
    Changelog:

- add missing docstrings for functions and classes;
- remove stray TODOs left in;
- remove a useless double loop;
- remove all "greedy" algorithms in favour of the Hungarian method;
- set "show_stats" in visualisation functions to False by default.

  • v2.2.2

    protected
    93bcc433 · Update package version · 
    Changelog:

- refactored classes in order to conform to the dataclass standard;
- specify positional vs. keyword-only arguments in most function signatures.

  • v2.2.1

    protected
    88194610 · Fix not exposing visualisation submodule · 
    This update cleans up package dependencies and moves to only using pyproject.toml to manage package metadata.

Further, a bug has been fixed: the package was not exposing the visualisation submodule.

  • v2.2.0

    protected
    ffd23af0 · Update package requirements · 
    This version includes a new, interactive visualisation which allows users to see which bounding boxes in a dataset contribute to agreement (or disagreement).

The package was also updated to be released in the future.

  • V2.1.3

    5c81b71a · Update package version · 
    This version sets the default matching algorithm for Kappa to `HUNGARIAN_EXCL_NULL`.

In future versions, `GREEDY` will be removed.

  • v2.1.2

    protected
    03b37b0d · Update package version · 
    This version fixes the percentage of mismatched bounding boxes being calculated in the wrong way.

This version also adds the main Alpha metrics onto the image stats.

  • v2.1.1

    protected
    ec22e008 · Update package version · 
    This version adds some things I forgot:

- added mean Alpha calculation;
- added safeguards for per-image Alpha calculation;
- added mismatched bounding box count to Result.

  • v2.1.0

    protected
    27a42e6a · Update package version · 
    This version adds the computation of Krippendorff's Alpha coefficient on a per-image basis, plus a number of improvements.