SmartLabeling

SmartLabeling 是一个 AI 辅助数据标注工具。它把桌面端交互、Python 推理服务和标注导出流程连在一起，用于降低基于图像 mask 的标注成本。

项目背景

传统图像标注需要大量手工描边。SmartLabeling 的核心问题是：如何让标注员用更少的点击表达意图，同时保留可撤销、可导出、可继续修正的工程流程。

系统结构

实现笔记

• Prompt operation state: foreground/background points and boxes are tracked with undo/redo history before regenerating masks.
• Async model orchestration: mask generation and class inference can run through Python service boundaries instead of blocking UI logic.
• Export discipline: generated masks and YAML label names are written as explicit project artifacts, not hidden UI state.
• Desktop packaging path: the repository includes Electron build configuration and keeps the Python `core/` service as part of the packaged file set.

过程中的取舍

• Local inference gives privacy and control, but increases installation and GPU-runtime complexity.
• Electron lowers desktop delivery cost, but requires clear separation between renderer interaction and Python service responsibilities.
• Mask quality should be presented honestly: the browser demo should show saved/static artifacts unless live inference is actually running.

当前状态

• The repository contains a Vue/Electron frontend, Python FastAPI service, SAM/CLIP model integration files, and desktop packaging configuration.
• The service layer includes prompt, project, image, mask, CLIP, and export modules, connecting the annotation path from interaction to saved artifacts.
• 目前还缺少有代表性的界面截图和 mask 示例，现阶段页面主要记录代码结构与流程。

相关入口

想继续了解实现时，可以从这些目录开始：

• Vue annotation components - 标注工具、内容区和提示输入的交互边界。
• FastAPI routers - prompt、upload、export、project switching 等服务入口。
• Python services - mask、CLIP、project 和 export 的业务实现。
• Desktop build config - Electron 打包和前端构建脚本。

后续记录

后续会补本地截图或 mask 示例到 `assets/projects/smartlabeling/`，再在本页加入静态 before/after 对比，把标注结果与使用过程记录得更完整。