Skip to content

v2.0.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 15 Sep 10:07
· 6 commits to main since this release
v2.0.0
2308f62

FaceSort 2.0.0

中文

随时暂停、随时继续，人工调整不会被覆盖

  • 整理途中点「暂停」：已认出的照片先归入各自文件夹，还没看到的暂存到 _待分类，进度自动保存；再按一次「立即停止」连暂存也停。
  • 对同一照片目录和输出目录再整理一次即可续跑：只处理剩下的照片，已归位的文件夹结构原样不动。中间手动把照片换到别的人物文件夹、或亲自把待分类里的照片归好，续跑都会尊重，不会复制回去。
  • 完成后 _待分类 自动清空消失。「整理设置」页会提示未完成的任务，预览只列出剩余部分。
  • 自动分组模式暂停时不写文件（分组依赖全量照片），已识别的结果已缓存，再点整理秒级接上。
  • 命令行：Ctrl-C 一次 = 暂停并暂存，两次 = 立即停止。

Windows 显卡加速，零配置

  • Windows 版改用 onnxruntime-directml：任何 DirectX 12 显卡（NVIDIA / AMD / Intel，含核显）自动加速，不用装 CUDA 或任何驱动工具包；没有可用显卡自动用 CPU。「设置」显示当前使用的加速器与回退原因。
  • 加速器与 CPU 的一致性由 CI 在 macOS（CoreML）和 Windows（DirectML）上逐次校验：embedding 余弦 ≥ 0.99、检测框一致。本机实测 CoreML 余弦 0.9998、框逐位相同，500ms → 60ms/张。

更快、更不重复

  • 识别缓存改为全局共享：换输出目录、改设置重跑、暂停后续跑，一张照片都不会被重新识别。「设置」可查看缓存占用并清除。
  • 添加样本时按缩放尺寸检查人脸，样本缩略图按需加载，样本多时启动不再变慢。

流程闭环与体验修复

  • 预览页与结果页现在会列出所有被跳过的文件和所有提示（样本无人脸、暂停、报告未写入…）。
  • 结果页按实际落盘计数；出错的照片逐条列出原因；「报告已保存」只在真的写入时显示。
  • 照片目录里之前的整理结果文件夹（如 _sorted）不再被当作输入重复整理。
  • 添加样本时若模型未安装，样本仍会保存并明确提示，不再静默下载。
  • 启动失败可重试；关闭窗口会立即停止后台整理并保存进度；Windows 路径显示与「打开文件夹」按钮文案按系统适配；自动分组识别中显示实时人脸计数；空目录有明确提示。

English

Pause and resume, without losing hand edits

  • Pause a sort: recognized photos go to their folders, the rest are parked in _待分类 (pending), progress is saved. Press again to stop even the parking.
  • Run the same folders again to resume: only pending photos are processed and the existing folder structure is left alone — including photos you moved by hand in between. _待分类 empties itself when done.
  • Auto-group mode pauses without writing files (groups need the whole shoot); the rerun finishes from cache in seconds.

GPU acceleration on Windows, zero setup

  • The Windows build now ships onnxruntime-directml: any DirectX 12 GPU (NVIDIA / AMD / Intel, integrated included) is used automatically, no CUDA toolkit; falls back to CPU otherwise. Settings shows what is active.
  • CI checks accelerator vs CPU agreement on macOS (CoreML) and Windows (DirectML).

Faster, never twice

  • One shared recognition cache across runs, output folders and settings changes. Visible and clearable in Settings.

Workflow fixes

  • Skipped files and warnings are shown on the preview and results pages; results count what actually landed on disk; per-file errors are listed; earlier result folders inside the input are skipped; adding samples without the model no longer downloads silently; closing the window stops the sort and saves progress.
Assets 4
Loading