360LayoutAnalysis - A lightweight open-source document layout analysis model based on YOLOv8.

## Overview of 360LayoutAnalysis 360LayoutAnalysis is a lightweight, open-source document layout analysis model developed by the 360 AI Research Institute. It is based on YOLOv8 and focuses on identifying and classifying document elements such as paragraphs, headings, images, and tables. The model is designed for applications like automated document processing, electronic data exchange, and historical document digitization. ## Key Features of 360LayoutAnalysis The key features of 360LayoutAnalysis include: - **Multi-scene support**: The model supports various document types, including Chinese papers, English papers, Chinese research reports, general scenarios, and textbook scenarios. - **Lightweight design**: Each model is only 6.23MB in size, making it suitable for environments with limited resources. - **Fine-grained label classification**: Each model provides specific label classifications for its target scenario, such as paragraphs, headings, images, and tables. ## Document Types Supported by 360LayoutAnalysis 360LayoutAnalysis supports the following document types: - Chinese papers - English papers - Chinese research reports - General scenarios - Textbook scenarios ## Overview of 360LayoutAnalysis 360LayoutAnalysis performs document layout analysis by detecting and classifying different elements within a document, such as paragraphs, headings, images, and tables. The model uses YOLOv8 to achieve this and is trained on high-quality datasets specific to each document type. ## Model Size in 360LayoutAnalysis Each model in 360LayoutAnalysis is lightweight, with a size of only 6.23MB. ## Usage of 360LayoutAnalysis 360LayoutAnalysis can be used by following these steps: 1. Ensure the ultralytics library is installed and the environment is correctly configured. 2. Specify the image path and model path (e.g., `paper-8n.pt` for Chinese papers). 3. Execute the prediction command, such as `result = model(image_path, save=True, conf=0.5, save_crop=False, line_width=2)`. 4. View the output results, which include detected bounding boxes, coordinates, class IDs, and confidence scores. ## License Information for 360LayoutAnalysis 360LayoutAnalysis is licensed under the Apache 2.0 license. For commercial use, users need to contact [email protected] for further details. ## Limitations of 360LayoutAnalysis The limitations of 360LayoutAnalysis include: - **Limited training data**: The model's training data is relatively limited, which may affect its generalizability compared to other models like DocLayout-YOLO. - **Specificity to certain scenarios**: While it performs well in specific scenarios, its performance may vary in more generalized or diverse document types. ### Citation sources: - [360LayoutAnalysis](https://huggingface.co/qihoo360/360LayoutAnalysis) - Official URL Updated: 2025-03-28