refactor: optimize openvino code

Author: SWHL

README.md

@@ -37,10 +37,9 @@ pip install rapid-layout onnxruntime
 ### 📋 使用
 
 ```python
-from rapid_layout import RapidLayout, RapidLayoutInput
+from rapid_layout import RapidLayout
 
-cfg = RapidLayoutInput()
-layout_engine = RapidLayout(cfg=cfg)
+layout_engine = RapidLayout()
 
 img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
 results = layout_engine(img_path)

demo.py

@@ -1,10 +1,11 @@
 # -*- encoding: utf-8 -*-
 # @Author: SWHL
 # @Contact: liekkaskono@163.com
-from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput
+from rapid_layout import EngineType, ModelType, RapidLayout
 
-cfg = RapidLayoutInput()
-layout_engine = RapidLayout(cfg=cfg)
+layout_engine = RapidLayout(
+    engine_type=EngineType.OPENVINO, model_type=ModelType.PP_LAYOUT_CDLA
+)
 
 img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
 results = layout_engine(img_path)

docs/install_usage/how_to_use_other_engine.md

@@ -7,54 +7,77 @@ hide:
 
 ## 引言
 
-因为版面分析模型输入图像尺寸固定，故可使用`onnxruntime-gpu`来提速。因为`rapid_layout`库默认依赖是CPU版`onnxruntime`，如果想要使用GPU推理，需要手动安装`onnxruntime-gpu`。详细使用和评测可参见[AI Studio](https://aistudio.baidu.com/projectdetail/8094594)
+版面分析支持多种推理引擎与设备：
 
-## 使用GPU
+- **ONNX Runtime**：默认引擎，支持 CPU / CUDA / DirectML / CANN，需按需安装对应包。
+- **OpenVINO**：可选，`pip install openvino` 后通过 `engine_type=EngineType.OPENVINO` 使用。
+
+默认依赖为 CPU 版 `onnxruntime`；使用 GPU 推理需手动安装 `onnxruntime-gpu`。详细使用和评测可参见 [AI Studio](https://aistudio.baidu.com/projectdetail/8094594)。
+
+## 使用 ONNX Runtime + GPU (CUDA)
 
 ```bash
 pip install rapid_layout
-
 # 请确保 onnxruntime-gpu 与当前 GPU/CUDA 版本对应
 # 参见 https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements
 pip install onnxruntime-gpu
 ```
 
-## 使用
-
 ```python linenums="1"
 from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput
 
 cfg = RapidLayoutInput(
     model_type=ModelType.PP_LAYOUT_CDLA,
     engine_type=EngineType.ONNXRUNTIME,
-    engine_cfg={"use_cuda": True, "cuda_ep_cfg.gpu_id": 1},
+    engine_cfg={"use_cuda": True, "cuda_ep_cfg": {"device_id": 0}},
 )
 layout_engine = RapidLayout(cfg=cfg)
 
 img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
 results = layout_engine(img_path)
 print(results)
+results.vis("layout_res.png")
+```
+
+多卡时可通过 `cuda_ep_cfg.device_id` 指定卡号（与 [engine_cfg.yaml](https://github.com/RapidAI/RapidLayout/blob/main/rapid_layout/configs/engine_cfg.yaml) 中 `cuda_ep_cfg.device_id` 一致）。
+
+## 使用 OpenVINO
+
+```bash
+pip install rapid-layout onnxruntime openvino
+```
+
+```python linenums="1"
+from rapid_layout import EngineType, ModelType, RapidLayout
 
+layout_engine = RapidLayout(
+    model_type=ModelType.PP_LAYOUT_CDLA,
+    engine_type=EngineType.OPENVINO,
+)
+img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
+results = layout_engine(img_path)
+print(results)
 results.vis("layout_res.png")
 ```
 
-## 使用NPU
+OpenVINO 设备与线程等配置见 [engine_cfg.yaml](https://github.com/RapidAI/RapidLayout/blob/main/rapid_layout/configs/engine_cfg.yaml) 中 `openvino` 段。
 
-详细配置参数参见：[engine_cfg.yaml](https://github.com/RapidAI/RapidLayout/blob/a7ab63ff291bd72e1a98ac2bb11860575514f432/rapid_layout/configs/engine_cfg.yaml)
+## 使用 NPU (CANN)
+
+详细配置参数参见：[engine_cfg.yaml](https://github.com/RapidAI/RapidLayout/blob/main/rapid_layout/configs/engine_cfg.yaml)
 
 ```python linenums="1"
 from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput
 
 cfg = RapidLayoutInput(
     model_type=ModelType.PP_LAYOUT_CDLA,
     engine_type=EngineType.ONNXRUNTIME,
-    engine_cfg={"use_cann": True, "cann_ep_cfg.gpu_id": 0},
+    engine_cfg={"use_cann": True, "cann_ep_cfg": {"device_id": 0}},
 )
 layout_engine = RapidLayout(cfg=cfg)
 
 img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
 results = layout_engine(img_path)
 print(results)
-
 results.vis("layout_res.png")
 ```

docs/install_usage/usage.md

@@ -5,28 +5,71 @@ hide:
   - toc
 ---
 
-
 ## Python 脚本运行
 
-```python
-from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput
+**默认用法**（默认模型 `pp_layout_cdla` + `onnxruntime` 引擎）：
 
-cfg = RapidLayoutInput()
-layout_engine = RapidLayout(cfg=cfg)
+```python
+from rapid_layout import RapidLayout
 
+layout_engine = RapidLayout()
 img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
 results = layout_engine(img_path)
 print(results)
+results.vis("layout_res.png")
+```
+
+**指定模型与引擎**（关键字参数）：
+
+```python
+from rapid_layout import EngineType, ModelType, RapidLayout
+
+layout_engine = RapidLayout(
+    model_type=ModelType.PP_LAYOUT_CDLA,
+    engine_type=EngineType.ONNXRUNTIME,
+    conf_thresh=0.5,
+    iou_thresh=0.5,
+)
+results = layout_engine(img_path)
+print(results)
+results.vis("layout_res.png")
+```
+
+**使用配置对象**（与上方等价）：
+
+```python
+from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput
 
+cfg = RapidLayoutInput(
+    model_type=ModelType.PP_LAYOUT_CDLA,
+    engine_type=EngineType.ONNXRUNTIME,
+    conf_thresh=0.5,
+    iou_thresh=0.5,
+)
+layout_engine = RapidLayout(cfg=cfg)
+results = layout_engine(img_path)
+print(results)
 results.vis("layout_res.png")
 ```
 
 ## 终端运行
 
 ```bash
 rapid_layout test_images/layout.png
+rapid_layout test_images/layout.png -m pp_layout_cdla --conf_thresh 0.5 --iou_thresh 0.5
 ```
 
+## 构造函数参数（RapidLayout / RapidLayoutInput）
+
+| 参数 | 类型 | 默认值 | 说明 |
+|------|------|--------|------|
+| `model_type` | ModelType / str | `pp_layout_cdla` | 模型类型 |
+| `model_dir_or_path` | str / Path / None | None | 模型路径，不传则按 model_type 解析 |
+| `engine_type` | EngineType / str | `onnxruntime` | 推理引擎：`onnxruntime`、`openvino` |
+| `engine_cfg` | dict | `{}` | 引擎额外配置 |
+| `conf_thresh` | float | 0.5 | 框置信度阈值 [0, 1] |
+| `iou_thresh` | float | 0.5 | IoU 阈值 [0, 1] |
+
 ## 可视化结果
 
 <div align="center">

docs/quickstart.md

@@ -12,29 +12,80 @@ hide:
 pip install rapid-layout onnxruntime
 ```
 
+如需使用 OpenVINO 引擎，请额外安装：`pip install openvino`。
+
 ## 运行
 
-=== "Python脚本运行"
+=== "Python 脚本（默认）"
+
+    不传参数时使用默认模型 `pp_layout_cdla` 与 `onnxruntime` 引擎：
 
     ```python linenums="1"
-    from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput
+    from rapid_layout import RapidLayout
 
-    cfg = RapidLayoutInput()
-    layout_engine = RapidLayout(cfg=cfg)
+    layout_engine = RapidLayout()
+    img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
+    results = layout_engine(img_path)
+    print(results)
+    results.vis("layout_res.png")
+    ```
+
+=== "Python 脚本（指定模型与引擎）"
+
+    通过关键字参数指定 `model_type`、`engine_type`、`conf_thresh` 等：
 
+    ```python linenums="1"
+    from rapid_layout import EngineType, ModelType, RapidLayout
+
+    layout_engine = RapidLayout(
+        model_type=ModelType.PP_LAYOUT_CDLA,
+        engine_type=EngineType.ONNXRUNTIME,
+        conf_thresh=0.5,
+    )
     img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
     results = layout_engine(img_path)
     print(results)
+    results.vis("layout_res.png")
+    ```
+
+=== "Python 脚本（使用配置对象）"
 
+    ```python linenums="1"
+    from rapid_layout import EngineType, ModelType, RapidLayout, RapidLayoutInput
+
+    cfg = RapidLayoutInput(
+        model_type=ModelType.PP_LAYOUT_CDLA,
+        engine_type=EngineType.ONNXRUNTIME,
+    )
+    layout_engine = RapidLayout(cfg=cfg)
+    img_path = "https://raw.githubusercontent.com/RapidAI/RapidLayout/718b60e927ab893c2fad67c98f753b2105a6f421/tests/test_files/layout.jpg"
+    results = layout_engine(img_path)
+    print(results)
     results.vis("layout_res.png")
     ```
 
 === "终端运行"
 
     ```bash linenums="1"
     rapid_layout test_images/layout.png
+    rapid_layout test_images/layout.png -m pp_layout_cdla --conf_thresh 0.5
     ```
 
+## 构造函数参数说明
+
+`RapidLayout(cfg=None, **kwargs)` 支持以下关键字参数（与 `RapidLayoutInput` 一致）：
+
+| 参数 | 类型 | 默认值 | 说明 |
+|------|------|--------|------|
+| `model_type` | `ModelType` 或 str | `pp_layout_cdla` | 模型类型，见 [模型列表](models.md) |
+| `model_dir_or_path` | str / Path / None | None | 模型路径，不传则按 `model_type` 自动解析 |
+| `engine_type` | `EngineType` 或 str | `onnxruntime` | 推理引擎：`onnxruntime`、`openvino` |
+| `engine_cfg` | dict | `{}` | 引擎额外配置，见 [engine_cfg.yaml](https://github.com/RapidAI/RapidLayout/blob/main/rapid_layout/configs/engine_cfg.yaml) |
+| `conf_thresh` | float | 0.5 | 框置信度阈值 [0, 1] |
+| `iou_thresh` | float | 0.5 | IoU 阈值 [0, 1] |
+
+传入 `cfg` 时，`kwargs` 会覆盖同名字段。
+
 ## 可视化结果
 
 <div align="center">

mkdocs.yml

@@ -146,7 +146,7 @@ markdown_extensions:
   - admonition
 
 nav:
-  - 概览: index.md
+  # - 概览: index.md
   - 快速开始: quickstart.md
   - 模型列表: models.md
   - 安装及使用:

rapid_layout/inference_engine/openvino/device_config.py

@@ -0,0 +1,48 @@
+# -*- encoding: utf-8 -*-
+# @Author: SWHL
+# @Contact: liekkaskono@163.com
+import os
+from typing import Any, Dict
+
+from omegaconf import DictConfig
+
+from ...utils.logger import logger
+
+
+class OpenVINOConfig:
+    def __init__(self, engine_cfg: DictConfig):
+        self.cfg = engine_cfg
+
+    def get_config(self) -> Dict[str, Any]:
+        config = {}
+
+        infer_num_threads = self.cfg.get("inference_num_threads", -1)
+        if infer_num_threads != -1 and 1 <= infer_num_threads <= os.cpu_count():
+            config["INFERENCE_NUM_THREADS"] = str(infer_num_threads)
+
+        performance_hint = self.cfg.get("performance_hint", None)
+        if performance_hint is not None:
+            config["PERFORMANCE_HINT"] = str(performance_hint)
+
+        performance_num_requests = self.cfg.get("performance_num_requests", -1)
+        if performance_num_requests != -1:
+            config["PERFORMANCE_HINT_NUM_REQUESTS"] = str(performance_num_requests)
+
+        enable_cpu_pinning = self.cfg.get("enable_cpu_pinning", None)
+        if enable_cpu_pinning is not None:
+            config["ENABLE_CPU_PINNING"] = str(enable_cpu_pinning)
+
+        num_streams = self.cfg.get("num_streams", -1)
+        if num_streams != -1:
+            config["NUM_STREAMS"] = str(num_streams)
+
+        enable_hyper_threading = self.cfg.get("enable_hyper_threading", None)
+        if enable_hyper_threading is not None:
+            config["ENABLE_HYPER_THREADING"] = str(enable_hyper_threading)
+
+        scheduling_core_type = self.cfg.get("scheduling_core_type", None)
+        if scheduling_core_type is not None:
+            config["SCHEDULING_CORE_TYPE"] = str(scheduling_core_type)
+
+        logger.info(f"Using OpenVINO config: {config}")
+        return config