Repolode
投稿する
BIT-DataLab

BIT-DataLab/Edit-Banana

Converts static diagram images into editable DrawIO (XML) files using SAM 3 segmentation, OCR, and LaTeX formula recognition.

GitHub で見る公式サイト
スター
5,498
フォーク
353
未解決の Issue
40
最終コミット
一昨日

これは何？

Edit Banana is a Python tool for turning fixed images of diagrams, flowcharts, technical schematics, and formulas into editable DrawIO XML. Each element is reconstructed as a separate shape, so it can be dragged, restyled, or relabeled in DrawIO after conversion.

The pipeline segments diagram elements with a fine-tuned SAM 3 model, detects text with local Tesseract OCR or PaddleOCR, and sends text and formula regions to Pix2Text to produce LaTeX. A multimodal LLM guides a multi-round extraction step, and the results are merged into a single DrawIO XML file. Input can be PNG, JPG, BMP, TIFF, or WebP.

The repository provides a CLI (main.py) for single or batch images and a FastAPI server (server_pa.py) with a /convert endpoint. Model weights for SAM 3 must be downloaded separately and configured in config.yaml. Support for DrawIO template import and local VLM deployment is listed on the roadmap as planned.

主な機能

  • 01Segments diagram elements with a fine-tuned SAM 3 model.
  • 02Recognizes text with local Tesseract OCR or PaddleOCR.
  • 03Converts text and formula regions to LaTeX using Pix2Text.
  • 04Outputs editable DrawIO (XML) files, with each element independently selectable.
  • 05Provides both a command-line interface and a FastAPI web endpoint.

ユースケース

  • Converting figures from papers or reports into editable DrawIO diagrams.
  • Recovering editable flowcharts or architecture diagrams from old screenshots.
  • Batch-converting scanned technical schematics into DrawIO files.
  • Extracting formulas from diagram images as LaTeX.

トピック

#diagram-conversion#ocr#drawio#sam#latex#computer-vision#python#image-to-xml#ai#data

よくある質問

Edit-Banana とは？
Converts static diagram images into editable DrawIO (XML) files using SAM 3 segmentation, OCR, and LaTeX formula recognition. Edit Banana is a Python tool for turning fixed images of diagrams, flowcharts, technical schematics, and formulas into editable DrawIO XML. Each element is reconstructed as a separate shape, so it can be dragged, restyled, or relabeled in DrawIO after conversion.
Edit-Banana は無料で使えますか？ライセンスは？
Edit-Banana は AGPL-3.0 ライセンスのオープンソースです。商用製品で使う前にライセンス条件を確認してください。
Edit-Banana は今もメンテナンスされていますか？
活発にメンテナンス中。Edit-Banana の最終コミットは 2026年10月17日 です。
Edit-Banana は何に使えますか？
Converting figures from papers or reports into editable DrawIO diagrams. Recovering editable flowcharts or architecture diagrams from old screenshots. Batch-converting scanned technical schematics into DrawIO files. Extracting formulas from diagram images as LaTeX.
Edit-Banana はどこで入手できますか？
ソースコードとインストール手順は GitHub にあります：https://github.com/BIT-DataLab/Edit-Banana

概要はプロジェクトの README をもとに AI が生成しています。GitHub の数値は毎日更新されます。

関連プロジェクト