Edit Banana converts static diagrams and PDFs into editable DrawIO or PPTX formats. The project leverages a fine-tuned SAM 3 model and multimodal large language models (LLMs) like Qwen-VL/GPT-4V for accurate element segmentation and reconstruction. It tackles the challenge of making traditionally uneditable content manipulable, preserving layout, styles, and text data.
Edit Banana distinguishes itself through its robust use of SAM 3 for accurate segmentation and a multi-stage process involving both vision and language models. The integration of both Tesseract OCR and Pix2Text addresses text recognition challenges, including formula conversion. The project supports a user-friendly interface and efficient resource management with a credit system and concurrency features, making it suitable for both individual users and potentially collaborative workflows.
- Advanced Segmentation: Utilizes SAM 3 for precise detection of diagram elements, enabling accurate reconstruction. <br> - Multi-Stage Extraction: Combines Tesseract OCR for local text and formula recognition with Pix2Text for enhanced understanding. <br> - Flexible Output: Supports conversion to both DrawIO (XML) and PPTX formats, catering to different user needs.<br> - User-Friendly Interface: Offers a command-line interface (CLI) and a user-friendly structure for easy integration and operation.<br> - Resource Management: Implements a credit system and LRU cache for efficient resource use and performance.
The project is actively under development, demonstrating a functional core pipeline with planned improvements like improved arrow connections and support for custom templates. Recent commits and active community engagement suggest ongoing maintenance and a commitment to feature enhancement, though the web-based service is recommended for the latest features.
Edit Banana benefits designers, analysts, and anyone working with static diagrams who need to make them editable. It addresses the limitations of traditional diagram formats by providing a tool to convert them into dynamic, manipulable assets. Users can leverage this for tasks like collaborative diagramming, data visualization enhancement, and creating interactive presentations, significantly expanding the utility of existing diagrams.