tsunghan-wu / d2ada Goto Github PK

Hi, I notice supervised learning on target domain only achieves 71.3% mIoU, which is much lower than proposed in mmsegmentation. Any reasons for such phenomenon?

Dear author,
thanks for your great work. I want to ask a simple question about visualization. For Fig.5 in the ablation study part, How to count the class frequency and selection changing rate and then plot them in the signal figure?

Hello there,

Thanks for sharing such a well-organized repo, it'll be very helpful for the active learning community.

However, I have some questions about the training set of supervised finetuning.

In line 45 of dataloader/active_dataset.py, why the training set of iteration 1 is still the source labeled set? Since in the end of the iteration 0, we have already queried some samples for target labeled set.

What's more, in RegionActiveDataset, the training set of supervised finetuning for every iteration is the concat of source labeled set and target labeled set. I'm confused of the inconsistency between ActiveDataset and RegionActiveDataset.

Thank you, and I look forward to hearing from you.