Requirements

Requirements

labelme support requires the image-dataset-converter-labelme library.

Image classification

The following converts all labelme .json files with image/label definitions into sub-directory format:

idc-convert -l INFO \
  from-labelme-ic \
    -l INFO \
    -i ./labelme/*.json \
  to-subdir-ic \
    -l INFO \
    -o ./subdir/

And this reads annotations in ADAMS .report format and stores in labelme json format and also specifies all possible labels:

idc-convert -l INFO \
  from-adams-ic \
    -l INFO \
    -i ./adams/*.report \
    -c classification \
  to-labelme-ic \
    -l INFO \
    --labels cat dog bunny bird \
    -o ./labelme

Object detection

The following converts ADAMS object annotations to labelme rectangle shapes:

idc-convert -l INFO \
  from-adams-od \
    -l INFO \
    -i ./adams/*.report \
  coerce-box \
  to-labelme-od \
    -l INFO \
    -o ./labelme

NB: coerce-box removes any polygon annotations and only leaves the bbox.

Instance segmentation

The following converts ADAMS object annotations to labelme polygon shapes:

idc-convert -l INFO \
  from-adams-od \
    -l INFO \
    -i ./adams/*.report \
  coerce-mask \
  to-labelme-od \
    -l INFO \
    -o ./labelme

NB: coerce-mask ensures that polygon annotations are present, even if it is only the outline of a bbox.