[Feat] Add PaddleOCRLoader for local OCR LangChain integration

Add PaddleOCRLoader that wraps the local PaddleOCR library (PP-OCRv5 and PP-StructureV3) to produce LangChain Document objects without requiring any cloud API.

New files:

- langchain_paddleocr/document_loaders/paddleocr.py: PaddleOCRLoader, PaddleOCRConfig dataclass, custom exception hierarchy

- tests/unit_tests/document_loaders/test_paddleocr_loader.py: 29 unit tests

- tests/integration_tests/document_loaders/test_paddleocr_loader.py: Integration tests

Modified files:

- langchain_paddleocr/__init__.py: Add PaddleOCRLoader export (lazy import for PaddleOCRVLLoader)

- langchain_paddleocr/document_loaders/__init__.py: Same

- README.md / README_cn.md: Add PaddleOCRLoader usage docs

Author: Ihebdhouibi

langchain-paddleocr/README.md

@@ -41,6 +41,63 @@ for doc in docs[:2]:
     print("---")
 ```
 
+### `PaddleOCRLoader`
+
+The `PaddleOCRLoader` wraps the **local** PaddleOCR library to extract text from PDF and image files — no cloud API or access token required.
+
+It supports two modes:
+
+- **Basic OCR** (default) — fast text extraction using PP-OCRv5.
+- **Structure mode** — layout-aware extraction (tables, titles, figures) using PP-StructureV3.
+
+#### Basic OCR
+
+```python
+from langchain_paddleocr import PaddleOCRLoader
+
+loader = PaddleOCRLoader(file_path="path/to/document.pdf")
+docs = loader.load()
+
+for doc in docs:
+    print(f"Page {doc.metadata['page']}: {doc.page_content[:100]}...")
+    print(f"Confidence: {doc.metadata['confidence']:.2f}")
+```
+
+#### Structure mode
+
+```python
+from langchain_paddleocr import PaddleOCRLoader
+from langchain_paddleocr.document_loaders.paddleocr import PaddleOCRConfig
+
+config = PaddleOCRConfig(lang="en", use_table_recognition=True)
+loader = PaddleOCRLoader(
+    file_path=["page1.png", "page2.png"],
+    use_structure=True,
+    config=config,
+)
+
+for doc in loader.lazy_load():
+    print(doc.page_content)
+    print(doc.metadata["layout_blocks"])
+```
+
+#### Configuration
+
+Use `PaddleOCRConfig` to pass engine parameters:
+
+| Parameter | Type | Description |
+|-----------|------|-------------|
+| `lang` | `str` | Language code (`"ch"`, `"en"`, `"fr"`, etc.) |
+| `ocr_version` | `str` | Pipeline version (`"PP-OCRv3"`, `"PP-OCRv4"`, `"PP-OCRv5"`) |
+| `use_doc_orientation_classify` | `bool` | Enable document orientation classification |
+| `use_doc_unwarping` | `bool` | Enable document de-warping |
+| `text_det_thresh` | `float` | Detection confidence threshold |
+| `text_rec_score_thresh` | `float` | Recognition confidence threshold |
+| `use_table_recognition` | `bool` | Enable table recognition (structure mode) |
+| `use_chart_recognition` | `bool` | Enable chart recognition (structure mode) |
+
+See the full list in `PaddleOCRConfig`.
+
 ## 📖 Documentation
 
 For full documentation, see the [API reference](https://reference.langchain.com/python/integrations/langchain_paddleocr/). For conceptual guides, tutorials, and usage examples, see the [LangChain Docs](https://docs.langchain.com/oss/python/integrations/providers/paddleocr).

langchain-paddleocr/README_cn.md

@@ -42,6 +42,63 @@ for doc in docs[:2]:
 ```
 
 
+### `PaddleOCRLoader`
+
+`PaddleOCRLoader` 封装了 **本地** PaddleOCR 库，从 PDF 和图像文件中提取文本 — 无需云 API 或访问令牌。
+
+支持两种模式：
+
+- **基础 OCR**（默认）— 使用 PP-OCRv5 进行快速文本提取。
+- **版面分析模式** — 使用 PP-StructureV3 进行版面感知提取（表格、标题、图片等）。
+
+#### 基础 OCR
+
+```python
+from langchain_paddleocr import PaddleOCRLoader
+
+loader = PaddleOCRLoader(file_path="path/to/document.pdf")
+docs = loader.load()
+
+for doc in docs:
+    print(f"页面 {doc.metadata['page']}: {doc.page_content[:100]}...")
+    print(f"置信度: {doc.metadata['confidence']:.2f}")
+```
+
+#### 版面分析模式
+
+```python
+from langchain_paddleocr import PaddleOCRLoader
+from langchain_paddleocr.document_loaders.paddleocr import PaddleOCRConfig
+
+config = PaddleOCRConfig(lang="ch", use_table_recognition=True)
+loader = PaddleOCRLoader(
+    file_path=["page1.png", "page2.png"],
+    use_structure=True,
+    config=config,
+)
+
+for doc in loader.lazy_load():
+    print(doc.page_content)
+    print(doc.metadata["layout_blocks"])
+```
+
+#### 配置
+
+使用 `PaddleOCRConfig` 传递引擎参数：
+
+| 参数 | 类型 | 说明 |
+|------|------|------|
+| `lang` | `str` | 语言代码（`"ch"`、`"en"`、`"fr"` 等） |
+| `ocr_version` | `str` | 流水线版本（`"PP-OCRv3"`、`"PP-OCRv4"`、`"PP-OCRv5"`） |
+| `use_doc_orientation_classify` | `bool` | 启用文档方向分类 |
+| `use_doc_unwarping` | `bool` | 启用文档去弯曲 |
+| `text_det_thresh` | `float` | 检测置信度阈值 |
+| `text_rec_score_thresh` | `float` | 识别置信度阈值 |
+| `use_table_recognition` | `bool` | 启用表格识别（版面分析模式） |
+| `use_chart_recognition` | `bool` | 启用图表识别（版面分析模式） |
+
+完整参数请参阅 `PaddleOCRConfig`。
+
 ## 📖 文档
 
 完整文档请参阅 [API 参考](https://reference.langchain.com/python/integrations/langchain_paddleocr/)。有关概念指南、教程和使用示例，请参阅 [LangChain 文档](https://docs.langchain.com/oss/python/integrations/providers/paddleocr)。

langchain-paddleocr/langchain_paddleocr/__init__.py

@@ -1,3 +1,12 @@
-from .document_loaders import PaddleOCRVLLoader
+from .document_loaders import PaddleOCRLoader
 
-__all__ = ["PaddleOCRVLLoader"]
+__all__ = ["PaddleOCRLoader", "PaddleOCRVLLoader"]
+
+
+def __getattr__(name: str) -> object:
+    if name == "PaddleOCRVLLoader":
+        from .document_loaders import PaddleOCRVLLoader
+
+        return PaddleOCRVLLoader
+    msg = f"module {__name__!r} has no attribute {name!r}"
+    raise AttributeError(msg)

langchain-paddleocr/langchain_paddleocr/document_loaders/__init__.py

@@ -1,3 +1,12 @@
-from .paddleocr_vl import PaddleOCRVLLoader
+from .paddleocr import PaddleOCRLoader
 
-__all__ = ["PaddleOCRVLLoader"]
+__all__ = ["PaddleOCRLoader", "PaddleOCRVLLoader"]
+
+
+def __getattr__(name: str) -> object:
+    if name == "PaddleOCRVLLoader":
+        from .paddleocr_vl import PaddleOCRVLLoader
+
+        return PaddleOCRVLLoader
+    msg = f"module {__name__!r} has no attribute {name!r}"
+    raise AttributeError(msg)