This project provides an integrated, cross-platform (Windows/Linux/Mac) open-source tool for image and video annotation, with a strong focus on streamlining the creation of datasets for YOLO object detection models. It targets engineers and researchers needing an efficient workflow from data labeling to model training, offering AI-powered automation to accelerate the process.

How It Works

Built with Python and Flask, xclabel functions as a web application accessible via a browser. Its core innovation lies in its dual capability: robust manual annotation tools (supporting various shapes and data import) and advanced AI-assisted auto-labeling. The latter integrates with large language models like LMStudio, vLLM, Ollama, and Alibaba Cloud services to automatically generate annotations for images and videos. The platform also encompasses a full YOLO model training pipeline, from dataset management to model export.

Quick Start & Requirements

Installation involves setting up a Python virtual environment and installing dependencies via pip. Key requirements include ultralytics==8.3.1, numpy==1.26.4, and specific torch/torchvision versions for CPU or CUDA (e.g., cu121). The service is started with python app.py --host 0.0.0.0 --port 9924, and accessed at http://127.0.0.1:9924. Model training is available at http://127.0.0.1:9924/training. All static resources are localized for offline deployment.

Highlighted Details

• AI auto-labeling supports multiple large model backends for automated image and video annotation.
• End-to-end YOLO model training pipeline includes dataset upload, training, breakpoint recovery, testing, and model download.
• YOLO format dataset export with customizable train/validation/test split ratios.
• Built-in file management system for browsing, uploading, and downloading files.
• Full offline deployment capability due to localized static assets.

Maintenance & Community

The project is maintained by beixiaocai. Links to the Gitee and GitHub repositories are provided. No specific community channels (like Discord/Slack) or detailed contributor information are highlighted in the README.

Licensing & Compatibility

The project's own code is released under the MIT license, allowing for free use, modification, and distribution, provided copyright notices are retained. Usage of third-party libraries is subject to their respective licenses. The MIT license generally permits commercial use and integration into closed-source projects.

Limitations & Caveats

The README does not explicitly detail limitations. However, users must ensure compatibility with the specified Python versions and specific versions of core dependencies like ultralytics and torch, particularly when opting for CUDA acceleration, which requires a compatible NVIDIA GPU and driver setup.