Meta-Dataset in TFDS - Train on Meta-Dataset episodes - md_sources list is too large?

[jfb54]

On the Meta-Dataset in TFDS documentation page (https://github.com/google-research/meta-dataset/blob/main/meta_dataset/data/tfds/README.md), in the "Train on Meta-Dataset episodes" section, there is a code example that says:

md_sources = ('aircraft', 'cu_birds', 'dtd', 'fungi', 'ilsvrc_2012', 'mscoco',
              'omniglot', 'quickdraw', 'traffic_sign'):
if md_version == 'v1':
  md_sources += ('vgg_flower',)

However, according to the VTAB+MD paper, the training datasets should not include 'mscoco' or 'traffic_sign'. I suppose that this is unlikely to be a problem as those two datasets do not have training splits, but it is a bit confusing to have them listed.

[vdumoulin]

Thank you for pointing this out! This is an oversight on our part, I will push out a correction shortly.

[vdumoulin]

Fixed via e12d61d.