BIT-DA / CAF

[TKDE 2023 ESI Highly Cited Paper] A Collaborative Alignment Framework of Transferable Knowledge Extraction for Unsupervised Domain Adaptation
MIT License
10 stars 3 forks source link

Details about dataset Visda-2017 #2

Open 23425Ning opened 1 week ago

23425Ning commented 1 week ago

Thanks for ur excellent work! I have some questions about the dataset Visda-2017. From ur "README.md", the visda2017 is only two parts which is exactly train and validation. So ur results on this datset is to use the train part as the source domain and the validation part as the target domain? And the test part is not involved in experiments? I'm newbie in the term of domain adaption, so I dont really get it how to operate on this dataset. Thanks for ur excellent work again! Looking for ur reply!

BinhuiXie commented 1 week ago

Thanks for your attention on our work.

Right! For some traditional domain adaptation benckmarks, like VisDA-2017, Office-31, and ImageCLEF, we follow previous evaluation setup, that is, reporting the results of target train set. However, for DomainNet dataset, there is test set for target domain, so we report the results of target test set.

Hope this does help, and please feel free to ask if I fail to make it clear.