1) Possibility of usage of IDD labels in DA.
Only the labels related to DA challenge can be used. E.g. For unsupervised DA, IDD labels cannot be used but IDD training images can be used. You can use the complete training dataset for supervised DA. (Instructions page updated accordingly)
2) Can multiple teams from same college participate.
3) Is there any rule for the team to be formed?
4) For DA, can we modify source datasets to fit our model?
Yes, make sure that you only use the samples we have allowed.
5) Could I use pseudo label method in unsupervised domain adaptation?
You can use any method. Use of any dataset outside permitted samples and labels is not allowed.
6) Procedure to register the team.
We have added an option to enter the team member's name (comma separated) . The team should consist of not more than 4 members.
7) Meaning of affiliation
Its your academic institution or organization.
8) May I use IDD detection datasets to semantic segmentation challenge?
No, you can only use allowed samples (or a sub-set from them) and allowed labels.
9) For DA, may I use a part of source data you given? Not use all of it.
Yes, you can use the subset from samples we have allowed.
10) For supervised DA, is it ok to train on source and fine-tune on target samples?
Yes, you can do that only for supervised DA (and provided target data for semi-supervised DA), make sure that you use only the samples allowed for training from source data.
Note that you need to be better than baselines for DA related challenges to be a winner.
Refer to the short summary on http://cvit.iiit.ac.in/autonue2021/challenge/index.html