[YDLuo-1]

Using a different OCR model currently requires command-line arguments or source changes. Add Automatic labeling settings (Ctrl+M) to select official v5/v6 model presets or local detection and recognition inference directories. The selection is saved in ppocrlabel.json for each image directory and restored when that directory is reopened.

The full OCR pipeline, detector, and recognizer switch together after the selected models load successfully. A failed load or settings save keeps the previous model set. Explicit command-line model arguments retain priority, new directories default to v5 mobile, and batch labeling continues to skip confirmed images.

Local model names are read from inference.yml; PyYAML is declared as a direct dependency. English and Chinese menu resources and usage instructions are updated. v6 models require compatible PaddleOCR, PaddleX, and PaddlePaddle versions.

Validation:

• All 9 unit/regression tests on this branch pass, covering directory isolation, local model identity, CLI precedence, settings write failure, atomic model switching, language selection, and the existing Windows trash behavior.
• Embedded English/Chinese menu resources match their source property files; Black and git diff --check pass.
• Windows CPU integration checks used Python 3.10, PaddlePaddle 3.1.1, PaddleOCR 3.7.0, and PaddleX 3.7.2 with official v5 mobile and locally exported v6 medium detection/recognition models. Model loading, real inference, directory restoration, failed-switch protection, and preservation of confirmed annotations were verified on a local dataset copy.

Public tests use synthetic configurations and temporary files. The v6 runtime check does not establish compatibility with every Paddle runtime or model preset.

[YDLuo-1]

其实今天还测试了调用deepseek模型帮忙标注，但是返回的框总是偏，效果不太行。之前有个数据集让codex帮忙标注效果挺好的。所以想看看能不能加个LLM自动标注的功能，效果不太行就没提PR，但感觉应该是可以搞的方向。有没有好用的提示词？