It is said that
"Unsupervised training: Models can be adapted (trained) on the test data in an unsupervised way, i.e. without labels."
However, it is also said that
"Can we assign pseudo labels to the unlabeled data in the target domain?
Assigning pseudo labels for the target training set is allowed as long as no human labeling is involved. Please DO NOT assign pseudo labels for the test set or validation."
Do they conflict with each other?
Posted by: madamada @ June 29, 2020, 4:13 a.m.Also confused about
"Unsupervised training: Models can be adapted (trained) on the test data in an unsupervised way, i.e. without labels."
and
"Can we use target_validation or target_test for training without using their labels?
No. Don't use target validation/test_set for training. only Target training set can be used for training."
Hi, thanks for your attention.
For the target domain, we provide the "target training set" for the model adaptation.
--> Unsupervised training: Models can be adapted (trained) on the test data in an unsupervised way, i.e. without labels.
This sentence is a little confusing. It should be:
--> Unsupervised training: Models can be adapted (trained) on the target domain (using the targe training set) in an unsupervised way, i.e. without labels.
Moreover, please do not use the "target test" and the "target validation" for the model adaptation. That is, assigning pseudo labels for the target training set is allowed as long as no human labeling is involved. Please DO NOT assign pseudo labels for the test set or validation."
We will update the first rule to make it clear. Thanks for pointing it out.
Posted by: weijian @ June 29, 2020, 4:30 a.m.Thanks for clarifying this. Could you send an email to everyone to inform the rules, otherwise some results on the leaderboard might be misleading.
Posted by: madamada @ June 29, 2020, 4:44 a.m.Thank you for your suggestion.
After the challenge ends, we will check the codes and require a short report to replicate their results. Thus, the final winning teams might not from the top teams in the leaderboard.
Posted by: weijian @ June 29, 2020, 4:52 a.m.