计算机视觉

本章节定义的所有类都将通过标准库cv提供。普通用户只需通过import方式导入cv标准库，无需自己写类别定义。

在下面的例子中，我们使用名为``MyClassDom``的类别域。

图像分类

图像分类任务是给每张图像分配一个类别标签。

类定义示例:

ImageClassificationSample:
    # Each sample contains an image together with a class label (optional).
    $def: struct
    $params: ['cdom'] 
    $fields:
        image: Image
        label: Label[dom=$cdom]
    $optional: ['label']

其中, $optional字段在数据样例中可以被省略。当$optional中label被忽略时，对应的字段值将被设置成一个空值（对应Python中的None）

数据示例:

data:
    sample-type: ImageClassificationSample[cdom=MyClassDom]
    sample-path: $local
    samples:
        - { image: "xyz/0001.jpg", label: "cat" }
        - { image: "xyz/0002.jpg", label: "dog" }
        - { image: "xyz/0003.jpg" }

目标检测

目标检测任务是检测给定图像中的有意义目标。每个被检测的目标可以被表示为一个带目标类别信息的边界框。

类定义示例:

LocalObjectEntry:
    # Each sample contains a bounding box together a class label (optional).
    $def: struct
    $params: ['cdom']
    $fields:
        bbox: BBox
        label: Label[dom=$cdom]
    $optional: ['label']

ObjectDetectionSample:
    # Each sample contains an image together with a list of local objects.
    $def: struct
    $params: ['cdom']
    $fields:
        image: Image
        objects: List[LocalObjectEntry[cdom=$cdom]]

数据示例:

data:
    sample-type: ObjectDetectionSample[cdom=MyClassDom]
    sample-path: $local
    samples:
        - image: "xyz/0001.jpg"
            objects: 
            - { bbox: [x1, y1, w1, h1], label: 1 }
            - { bbox: [x2, y2, w2, h2], label: 2 }
            - { bbox: [x3, y3, w3, h3], label: 2 }
        - image: "xyz/0002.jpg"
            objects: 
            - { bbox: [x4, y4, w4, h4], label: 3 }
            - { bbox: [x5, y5, w5, h5], label: 4 }

带场景分类的目标检测

在某些应用中，场景分类（图像级别）及目标检测会组成一个联合任务。对于这样一个任务，我们有两个类别域，这里称为SceneDom 和 ObjectDom，分别限定场景类别和目标类别。

类定义示例:

SceneAndObjectSample:
    # Each samples contains an image, a scene label (optional)
    # together with a list of local objects.
    $def: struct
    $params: ['scenedom', 'objectdom']
    $fields:
        image: Image 
        sclabel: Label[dom=$scenedom]
        objects: List[LocalObjectEntry[cdom=$objectdom]]
    $optional: ['sclabel']

数据示例:

data:
    sample-type: SceneAndObjectSample[scenedom=SceneDom, objectdom=ObjectDom]
    sample-path: $local
    samples:
        - image: "xyz/0001.jpg"
            sclabel: "street"
            objects: 
            - { bbox: [x1, y1, w1, h1], label: 1 }
            - { bbox: [x2, y2, w2, h2], label: 2 }
            - { bbox: [x3, y3, w3, h3], label: 2 }
        - image: "xyz/0002.jpg"
            sclabel: "river"
            objects: 
            - { bbox: [x4, y4, w4, h4], label: 3 }
            - { bbox: [x5, y5, w5, h5], label: 4 }

图像分割

图像分割任务是对给定图像中的每个像素点指定像素级标签。惯例是使用标签图，也就是在一个额外的文件中存储一个非结构化目标。

类定义示例:

ImageSegmentationSample:
    # Each sample contains an image together with a label map.
    $def: struct
    $params: ['cdom']
    $fields:
        image: Image
        labelmap: LabelMap[dom=$cdom]

数据示例:

data:
    sample-type: ImageSegmentationSample[cdom=MyClassDom]
    sample-path: $local
    samples:
        - { image: "imgs/0001.jpg", labelmap: "maps/0001.ppm" }
        - { image: "imgs/0002.jpg", labelmap: "maps/0002.ppm" }
        - { image: "imgs/0003.jpg", labelmap: "maps/0003.ppm" }