EXTRACT_BOUNDING_BOX
ExtractBoundingStep is a a pipeline step that extracts sub-images from an input image, based on the locations of input bounding boxes. Returns list of image for the cropped image regions.
Last updated
ExtractBoundingStep is a a pipeline step that extracts sub-images from an input image, based on the locations of input bounding boxes. Returns list of image for the cropped image regions.
Last updated
Configs
Descriptions
imageName
Name of the input image key from the previous step. If set to null, it will try to find any image in the incoming data instance.
bboxName
Name of the bounding boxes key from the previous step. If set to null, it will try to find any bounding box in the incoming data instance.
outputName
Name of the output key that will contain the output as images the input image region covered by the bounding boxes.
keepOtherFields
If true, other data key and values from the previous step are kept and passed on to the next step as well. Default is set to true.
aspectRatio
If set, the smaller dimensions will be increased to keep the aspect ratio correct which may crop outside the image border.
resizeH
If specified, the cropped images will be resized to the specified height.
resizeW
If specified, the cropped images will be resized to the specified width.
imageToNDArrayConfig
Used to account for the fact that n-dimensional array from ImageToNDArrayConfig may be used to crop images before passing to the network, when the image aspect ratio does not match the NDArray aspect ratio. This allows the step to determine the subset of the image actually passed to the network that produced the bounding boxes. Refer config from IMAGE_TO_NDARRAY.