segmentation

Type Members

1. case class COCOAnotationOD(id: Long, imageId: Long, categoryId: Long, segmentation: COCOSegmentation, area: Float, bbox: (Float, Float, Float, Float), isCrowd: Boolean, image: COCOImage = null) extends Product with Serializable

   

   An annotation for an image (OD in the name for Object Detection)

   An annotation for an image (OD in the name for Object Detection)

   imageId

   the Id of the image

   categoryId

   the Id of the category. Note that categoryId is not continuous from 0 to the number of categories. You can use COCODataset.cateId2Idx to convert an categoryId to a compact category index.

   segmentation

   the segmentation data

   area

   area

   bbox

   the bounding box, (xmin, ymin, xmax, ymax)

   isCrowd

   if the annotation is a crowd. e.g. a crowd of people. If true, segmentation is an COCORLE object

   image

   the reference to the image

2. case class COCOCategory(id: Long, name: String) extends Product with Serializable

   

3. case class COCODataset(info: COCODatasetInfo, images: Array[COCOImage], annotations: Array[COCOAnotationOD], licenses: Array[COCOLicence], categories: Array[COCOCategory]) extends Product with Serializable

   

4. case class COCODatasetInfo(year: Int, version: String, description: String, contributor: String, url: String) extends Product with Serializable

   

5. case class COCOImage(id: Long, height: Int, width: Int, license: Int) extends Product with Serializable

   

6. case class COCOLicence(id: Long, name: String, url: String) extends Product with Serializable

   

7. case class COCOPoly(_poly: Array[Array[Float]], _height: Int, _width: Int) extends PolyMasks with COCOSegmentation with Product with Serializable

   

8. case class COCORLE(_counts: Array[Int], _height: Int, _width: Int) extends RLEMasks with COCOSegmentation with Product with Serializable

   

9. trait COCOSegmentation extends AnyRef

   

10. class PolyMasks extends SegmentationMasks

    

    A mask of regions defined by one or more polygons.

    A mask of regions defined by one or more polygons. The masked object(s) should have the same label.

11. class RLEMasks extends SegmentationMasks

    

    A mask of regions defined by RLE.

    A mask of regions defined by RLE. The masked object(s) should have the same label. This class corresponds to "uncompressed RLE" of COCO dataset. RLE is a compact format for binary masks. Binary masks defines the region by assigning a boolean to every pixel of the image. RLE compresses the binary masks by instead recording the runs of trues and falses in the binary masks. RLE is an array of integer. The first element is the length of run of falses staring from the first pixel. The second element of RLE is the is the length of first run of trues. e.g. binary masks: 00001110000011 RLE: ---4--3----5-2 ====> 4,3,5,2

    Also note that we don't use COCO's "compact" RLE string here because this RLE class has better time & space performance.

12. abstract class SegmentationMasks extends Serializable