Supports lumi dataset transform --class-examples N, which makes the dataset construction finish when every class has at least N number of samples. This will be the attempted lower bound; more examples might be added or a class might finish with fewer samples depending on the dataset.
Removed useless "random classes" options & seed.
Fixed OpenImages reader.
Now works with correct folder structure.
Skips group annotations.
Print dataset per-class composition after any transform command.
lumi dataset transform --class-examples N
, which makes the dataset construction finish when every class has at leastN
number of samples. This will be the attempted lower bound; more examples might be added or a class might finish with fewer samples depending on the dataset.transform
command.