config.yaml

# ------------------------------------------------------------------------------

frontend:
  type: sparse
  image: [root_path]/sfs-framework-example/image
  mask: [root_path]/sfs-framework-example/mask.png
  scale: 0.5
  step: 1

features:
  threshold: 0.0005

matching:
  range: 5

algorithm:
  group: 3
  error: 1e-5
  disparity: 5
  radius: 50

dense:
  disparity: 3.3245e-04

export:
  group: 3
  path: [root_path]/sfs-framework-example/output

# ------------------------------------------------------------------------------
#
# frontend:
#
#   type:
#   This can be 'sparse' or 'dense' and is the framework mode. The 'sparse' has
#   to be computed first.
#
#   image:
#   Path of the image folder.
#
#   mask:
#   Path of the mask image (discarding un-relevant pixel areas).   
#
#   scale:
#   Scale factor of the image. Specifying 0.5 reduce the size of the image by
#   two.
#
#   step:
#   Image progression step. For example, specifying two leads the framework to
#   discard one image over two.
#
# features:
#
#   threshold:
#   OpenCV AKAZE features detection threshold. Decrease this value leads to
#   increase the amount of detected features.
#
# matching:
#
#   range:
#   Value specifying how many previous images are matched with the one added to
#   the reconstruction.
#
# algorithm:
#
#   group:
#   This value has to be at least three or greater. It is the amount of image,
#   counting the last one, used to propagate the scale factor as a new image
#   is added to the reconstruction.
#
#   error:
#   Error value used to stop the algorithm iteration. It is the value, according
#   to the mean distance between the images, below which the error is considered
#   sufficiently small.
#
#   disparity:
#   Standard deviation factor. This value indicates to the filtering which
#   tolerance is applied to keep or remove a link between a 3D point and an
#   image that sees it.
#
#   radius:
#   This value, factor of the mean distance between the images, fix the maximum
#   distance a 3D point can be from the images that see it.
#
# dense:
#
#   disparity:
#   Filtering force. As most of the image pixels are placed in the 3D space, a
#   filtering process is applied to keep only inliers. The re-projection error
#   is considered and applied on the unit sphere using this value as a maximum
#   tolerance.
#
# export
#
#   group:
#   Specify how many images a point has to be seen from to be exported in the
#   output model.
#
#   export:
#   Framework exportation path.
#
# ------------------------------------------------------------------------------