feat: LEAP-1937: Add brush_to_coco format

[KMYsqHRK]

Overview
This PR adds support for converting brush annotations (RLE-encoded) to COCO format. This enables using brush segmentation data in machine learning pipelines that require COCO-formatted annotations.
Changes

Added BRUSH_TO_COCO format to the converter module
Created brush_to_coco.py for RLE decoding and contour extraction
Updated converter interface and CLI options

Features

Converts RLE-encoded brush masks to COCO polygons
Handles mixed annotation types (brush, polygon, rectangle)
Preserves label categories and annotation metadata
Compatible with existing converter workflow

[KMYsqHRK]

Added OpenCV dependencies to pyproject.toml
Added pytest (tests/test_brush_to_coco.py)
fix bugs and update stability

Please review it !

[nikitabelonogov]

Hi @KMYsqHRK !
Thank you for your contribution!

Unfortunately this repository source code is automatically generated via Fern from https://github.com/HumanSignal/label-studio-client-generator/

your changes will be overridden with next regeneration

[nikitabelonogov]

Oh OK, nevermind. As for @makseq this part of code is not under auto generator.
I've created a mirror PR #432 for testing purposes.

[KMYsqHRK]

整数である必要があるようなので元々のコードではintで丸めていた部分をroundにしました。 テストの結果はほとんど変わらなかったのですが、とりあえずroundのままにしています。 2025年3月19日(水) 21:30 NAKAYAMA Atsushi ***@***.***>:

…

***@***.**** commented on this pull request. ------------------------------ In src/label_studio_sdk/converter/exports/brush_to_coco.py <#429 (comment)> : > + + +def generate_contour_from_polygon(points, original_width, original_height): + """Convert polygon annotations to COCO segmentation format + + Args: + points: List of [x, y] points in percentage format + original_width: Width of the original image + original_height: Height of the original image + + Returns: + tuple: (segmentation, bbox, area) in COCO format + """ + # Convert from percentage to absolute coordinates + modified_points = [ + [round(x * original_width / 100), round(y * original_height / 100)] round()で丸めてしまって大丈夫でしょうか? — Reply to this email directly, view it on GitHub <#429 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BGCF5YOG7R5NFPAIZSLLXS32VFPM3AVCNFSM6AAAAABY3J3QHOVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDMOJYGIZTINJQHA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[makseq]

Could you please update your branch with the latest master branch from the main LS SDK repo? I am trying to run tests for the main LS and they fail; I think it's because your branch is slightly outdated:
https://github.com/HumanSignal/label-studio/actions/runs/13946116271/job/39033643447?pr=7252

[KMYsqHRK]

Regarding the PR review, I'm a student part-timer currently writing a paper, so my response might be delayed. I apologize for the inconvenience!