本版本重点
v1.3.0 新增实验性的手动标注修复，解决自动检测漏掉黑条或选区不完整时无法补画的问题。对应 #10，已通过 #11 合入。
手动标注修复（实验性功能）
- 入口：单图片模式 + 模式 I → 高级功能（实验性功能）→ 手动标注修复。
- 可以先自动检测，再用画笔补画、橡皮擦除，按当前选区执行修复。
- 支持撤销/重做、缩放/平移、选区扩大预览和工作图/结果对比。
- 使用原图尺寸的独立遮罩，保留选区外像素及透明通道；另存结果时自动避让已有文件。
- 修复展开高级功能后的窗口适配，以及编辑器返回后的字体和字号一致性。
Warning
该手动涂抹功能除了能跑以外特别难用，这个项目也终于到需要重构的阶段了
目前仅支持静态单图片和模式 I；批量审核与整体重构留待后续处理。
文档
- 重新整理中文 README，补充图标、界面截图、安装步骤和操作说明。
- 新增 English README，顶部可切换语言；两个发行包都包含双语 README 和配套图片。程序界面目前仍为中文。
下载选择
Aletheia-Lens_v1.3.0_cpu.7z：CPU 版，适合没有兼容 NVIDIA 显卡的设备，下载体积较小。Aletheia-Lens_v1.3.0_cuda12.7z：NVIDIA GPU 版，内置 CUDA 12 / cuDNN 9 运行库。
两个压缩包均包含运行模型，解压后运行 Aletheia-Lens.exe，无需安装 Python 或另行下载模型。CUDA12 版优先让检测与 ESRGAN 放大使用 GPU；DeepCreamPy 的 bar.onnx / mosaic.onnx 因兼容性原因仍固定使用 CPU。
验证
- 本地及 CPU / CUDA12 CI 的 109 项回归测试通过。
- 本机图形界面和真实 CPU 手动修复验证通过，包括窗口适配、字体恢复、保存及延迟任务处理。
- 两个版本均通过模型 SHA-256 校验、PyInstaller 打包及打包后四模型真实推理检查；CUDA12 包额外通过 DLL 清单和 provider 检查。
- GitHub Windows runner 没有 NVIDIA GPU，上述云端检查不代表真实 GPU 执行已验证。
English
v1.3.0 adds an experimental manual mask editor for correcting missed or incomplete bar detections (#10, implemented in #11).
- Open Single image + Mode I → “高级功能（实验性功能）” → “手动标注修复”. Run automatic detection or paint a selection, erase false positives, then repair using the current mask.
- Includes undo/redo, zoom/pan, selection expansion preview, result comparison, and saving without overwriting the source or previous results.
- Preserves pixels outside the selection and the alpha channel. Fixes window sizing and font consistency when returning from the editor.
- Adds an English README and bilingual navigation. Both packages include the documentation and images; the application interface remains in Chinese.
Warning
The manual painting feature runs, but it is very difficult to use. This project has finally reached the point where it needs a refactor.
Only static single images and mode I are supported. Batch review and broader refactoring are deferred.
Downloads: choose *_cpu.7z for CPU inference or *_cuda12.7z for a compatible NVIDIA GPU. Both include all runtime models; extract and run Aletheia-Lens.exe. The CUDA12 package bundles CUDA 12 / cuDNN 9, while the DeepCreamPy bar and mosaic models remain pinned to CPU for compatibility.
Validation: 109 regression tests passed locally and in each CI build. Both packaged executables passed real inference checks for all four models; the CUDA12 package also passed DLL and provider checks. The desktop/manual repair workflow was verified locally on CPU. GitHub's Windows runners have no NVIDIA GPU, so cloud validation does not establish actual GPU execution.