Why use the target val datasets to evaluate the trained model？

Junjue-Wang / LoveDA

[NeurIPS 2021] LoveDA: A Remote Sensing Land-Cover Dataset for Domain Adaptive Semantic Segmentation

361 stars 50 forks source link

Why use the target val datasets to evaluate the trained model？ #81

Open StuLiu opened 1 year ago

StuLiu commented 1 year ago

Nice work! In the UDA setting, the labels of the target domain cannot be used to choose hyper-parameters. Why use the target val datasets to evaluate the trained model？