add yolo11 obb documentation

docs/doc/en/sidebar.yaml

@@ -75,6 +75,8 @@ items:
         label: Object tracking and counting
     -   file: vision/ocr.md
         label: OCR
+    -   file: vision/detect_obb.md
+        label: Detect with angle(OBB)
     -   file: vision/maixhub_train.md
         label: MaixHub online AI training
     -   file: vision/customize_model_yolov5.md

docs/doc/en/vision/detect_obb.md

@@ -0,0 +1,66 @@
+---
+tite: Target Detection with Rotation Angles (OBB, Oriented Bounding Box)
+update:
+    - date: 2024-12-20
+      version: v1.0
+      author: neucrack
+      content:
+        Support YOLO11/YOLOv8 OBB model and add documentation
+---
+
+## Introduction
+
+In standard object detection, the output is typically a rectangular bounding box. However, in certain scenarios, objects may have a rotated shape, requiring a bounding box with a rotation angle. This type of bounding box is referred to as an OBB (Oriented Bounding Box).  
+![](../../assets/ships-detection-using-obb.jpeg)
+
+**Standard detection results:** `x, y, w, h` represent the top-left or center coordinates of the rectangle, along with its width and height.  
+**OBB detection results:** `x, y, w, h, angle` includes an additional parameter for the rotation angle.
+
+## Using Rotated Bounding Boxes (OBB) in MaixPy MaixCAM
+
+`MaixPy` supports the `YOLO11/YOLOv8` OBB model, enabling quick and convenient implementation. Below is an example based on the `nn_yolo11_obb.py` script from [MaixPy/examples](https://github.com/sipeed/maixpy):
+
+```python
+from maix import camera, display, image, nn, app
+
+detector = nn.YOLO11(model="/root/models/yolo11n_obb.mud", dual_buff=True)
+
+cam = camera.Camera(detector.input_width(), detector.input_height(), detector.input_format())
+disp = display.Display()
+
+while not app.need_exit():
+    img = cam.read()
+    objs = detector.detect(img, conf_th=0.5, iou_th=0.45)
+    for obj in objs:
+        points = obj.get_obb_points()
+        msg = f'{detector.labels[obj.class_id]}: {obj.score:.2f}, {obj.angle * 180:.1f}'
+        img.draw_string(points[0], points[1] - 4, msg, color=image.COLOR_RED)
+        detector.draw_pose(img, points, 8 if detector.input_width() > 480 else 4, image.COLOR_RED, close=True)
+    disp.show(img)
+```
+
+In this example, the `YOLO11` model is used to load an OBB model. After detecting an object, the rotation angle can be accessed via `obj.angle`. The `x, y, w, h` attributes of `obj` represent the unrotated rectangle. The `get_obb_points` method retrieves the four vertices of the rotated rectangle, and `draw_pose` is used to draw the rotated bounding box. The `close` parameter ensures the vertices are connected.
+
+The default model is the official YOLO11 15-class model with the following labels:
+
+```python
+plane, ship, storage tank, baseball diamond, tennis court, basketball court, ground track field, harbor, bridge, large vehicle, small vehicle, helicopter, roundabout, soccer ball field, swimming pool
+```
+
+The model file can be found at `/root/models/yolo11n_obb.mud`.
+
+## Customizing Your Own OBB Model for MaixPy MaixCAM
+
+### Using the Model on a Computer
+
+For an introduction to the official `YOLO11` OBB models, refer to [YOLO11 OBB](https://docs.ultralytics.com/tasks/obb/). This documentation explains how to use OBB models on a computer and export ONNX model files.
+
+### Exporting the Model for MaixCAM
+
+Follow the [YOLO11/YOLOv8 Custom Models](./customize_model_yolov8.md) guide to convert an ONNX model into a MUD model compatible with MaixCAM.  
+**Note:** Ensure the output names of the model's layers are correctly configured during conversion.
+
+### Training Your Own OBB Model
+
+Refer to the [YOLO11 Official Training Documentation](https://docs.ultralytics.com/datasets/obb/dota-v2/) to prepare your dataset and train your own OBB model.
+

docs/doc/zh/sidebar.yaml

@@ -75,6 +75,8 @@ items:
         label: 物体轨迹跟踪和计数
     -   file: vision/ocr.md
         label: OCR 文字识别
+    -   file: vision/detect_obb.md
+        label: 带旋转角度的检测(OBB)
     -   file: vision/maixhub_train.md
         label: MaixHub 在线训练 AI 模型
     -   file: vision/customize_model_yolov5.md

docs/doc/zh/vision/customize_model_yolov8.md

@@ -68,7 +68,7 @@ model = YOLO(net_name)  # load an official model
 
 # Predict with the model
 results = model("https://ultralytics.com/images/bus.jpg")  # predict on an image
-path = model.export(format="onnx", imgsz=[input_height, input_width])  # export the model to ONNX format
+path = model.export(format="onnx", imgsz=[input_height, input_width], dynamic=False, simplify=True, opset=17)   # export the model to ONNX format
 print(path)
 
 ```
@@ -82,7 +82,7 @@ MaixPy/MaixCDK 目前支持了 YOLOv8 / YOLO11 检测 以及 YOLOv8-pose / YOLO1
 
 按照[MaixCAM 模型转换](../ai_model_converter/maixcam.md) 进行模型转换。
 
-注意模型输出节点的选择（注意可能你的模型可能数值不完全一样，看下面的图找到相同的节点即可）：
+注意模型输出节点的选择（**注意可能你的模型可能数值不完全一样，看下面的图找到相同的节点即可**）：
 * 检测模型：
   * YOLOv8 提取 onnx 的 `/model.22/dfl/conv/Conv_output_0,/model.22/Sigmoid_output_0` 这两个输出。
   * YOLO11 提取`/model.23/dfl/conv/Conv_output_0,/model.23/Sigmoid_output_0`输出。
@@ -92,8 +92,11 @@ MaixPy/MaixCDK 目前支持了 YOLOv8 / YOLO11 检测 以及 YOLOv8-pose / YOLO1
 * 图像分割：
   * YOLOv8-seg 提取 `/model.22/dfl/conv/Conv_output_0,/model.22/Sigmoid_output_0,/model.22/Concat_output_0,output1`
   * YOLO11-seg 提取 `/model.23/dfl/conv/Conv_output_0,/model.23/Sigmoid_output_0,/model.23/Concat_output_0,output1`四个输出。
+* OBB 检测：
+  * YOLOv8 提取`/model.22/dfl/conv/Conv_output_0,/model.22/Sigmoid_1_output_0,/model.22/Sigmoid_output_0`这三个输出。
+  * YOLO11 提取`/model.23/dfl/conv/Conv_output_0,/model.23/Sigmoid_1_output_0,/model.23/Sigmoid_output_0`这三个输出。
 
-![](../../assets/yolov8_out1.jpg) ![](../../assets/yolov8_out2.jpg)
+![](../../assets/yolov8_out1.jpg) ![](../../assets/yolov8_out2.jpg) ![](../../assets/yolo11_out_obb.jpg)
 
 对于物体检测，mud 文件为（YOLO11 model_type 改为 yolo11）
 ```ini

docs/doc/zh/vision/detect_obb.md

@@ -0,0 +1,69 @@
+---
+tite: 带旋转角度的目标检测(OBB, Oriented Bounding Box)
+update:
+    - date: 2024-12-20
+      version: v1.0
+      author: neucrack
+      content:
+        支持 YOLO11/YOLOv8 OBB 模型并添加文档
+---
+
+## 简介
+
+普通的检测中输出结果是一个矩形框，但是在一些场景中，目标物体的形状是旋转的，这时候就需要输出一个带旋转角度的矩形框，这种矩形框叫做OBB(Oriented Bounding Box)。
+![](../../assets/ships-detection-using-obb.jpeg)
+
+普通检测结果： x, y, w, h, 分别是 矩形框左上角或者中心点坐标，以及矩形宽高。
+OBB 检测结果：x, y, w, h, angle, 多了一个旋转角度。
+
+## MaixPy MaixCAM 中使用带旋转角度的目标检测(OBB)
+
+`MaixPy` 中移植了 `YOLO11/YOLOv8` `OBB` 模型，可以快速方便地实现， 以下为例程，以[MaixPy/examples](https://github.com/sipeed/maixpy)中的`nn_yolo11_obb.py`为准：
+```python
+from maix import camera, display, image, nn, app
+
+detector = nn.YOLO11(model="/root/models/yolo11n_obb.mud", dual_buff = True)
+
+cam = camera.Camera(detector.input_width(), detector.input_height(), detector.input_format())
+disp = display.Display()
+
+while not app.need_exit():
+    img = cam.read()
+    objs = detector.detect(img, conf_th = 0.5, iou_th = 0.45)
+    for obj in objs:
+        points = obj.get_obb_points()
+        msg = f'{detector.labels[obj.class_id]}: {obj.score:.2f}, {obj.angle * 180:.1f}'
+        img.draw_string(points[0], points[1] - 4, msg, color = image.COLOR_RED)
+        detector.draw_pose(img, points, 8 if detector.input_width() > 480 else 4, image.COLOR_RED, close=True)
+    disp.show(img)
+```
+
+可以看到这里使用`YOLO11`加载了`obb`模型，然后在检测到目标后，通过`obj.angle`获取到矩形旋转角度，`obj`的`x,y,w,h`属性是未旋转的矩形，通过`get_obb_points`获取到旋转后的矩形的四个顶点坐标，然后通过`draw_pose`绘制出目标的旋转矩形框，`close`参数表示将矩形框四个顶点连线起来。
+
+
+默认的模型是 YOLO11 官方的`15`类模型，分类标签如下：
+```python
+plane, ship, storage tank, baseball diamond, tennis court, basketball court, ground track field, harbor, bridge, large vehicle, small vehicle, helicopter, roundabout, soccer ball field, swimming pool
+```
+在`/root/models/yolo11n_obb.mud`文件中也能看到。
+
+
+## MaixPy MaixCAM 自定义自己的OBB模型
+
+### 电脑端使用模型
+
+关于 `YOLO11` 官方的 OBB 模型介绍请看[YOLO11 OBB](https://docs.ultralytics.com/tasks/obb/)。
+在这个文档中可以看到如何在电脑端使用`obb`模型，以及如何导出 ONNX 模型文件。
+
+### 导出模型给 MaixCAM 使用
+
+根据 [YOLO11/YOLOv8 自定义模型](./customize_model_yolov8.md) 即可将 ONNX 模型转换为 MaixCAM 可以使用的 MUD 模型。
+注意：转换时需要注意输出层的输出名不要弄错。
+
+### 训练自己的 OBB 模型
+
+根据[YOLO11 官方训练文档](https://docs.ultralytics.com/datasets/obb/dota-v2/) 准备自己的数据集，然后进行训练即可。
+
+
+
+