search

TKKim93 / APE

[ECCV 2020] Attract, Perturb, and Explore: Learning a Feature Alignment Network for Semi-supervised Domain Adaptation
30 stars 3 forks source link

varying number of labeled target samples per class #2

Closed zhaoxin94 closed 3 years ago

zhaoxin94 commented 4 years ago

Hi, your work is really interesting and impressive. Thank you for make your code open source!

In the Analysis section and supplementary material, you also did some experiments on 5(10, 15, 20)-shot settings. Do you mind to share your split (the txt file) on 5, 10, 15, 20 shot settings? It will help me a lot to reproduce your results on these settings.

TKKim93 commented 4 years ago

Thank you for visiting my repository!

I update the DomainNet splits for additional n-shot settings in the readme.

I wish it would help.

meghbhalerao commented 4 years ago

Hi @TKKim93, Thanks a lot for providing the lists for the 5,10,15 and 20 shot settings! Really appreciate it! I had a small question regarding it. Are you doing the validation with 3 examples per class, even with the 5,10,15 and 20 shot settings? Thanks, Megh

TKKim93 commented 4 years ago

Yes, I used the same split across all the n-shot setting experiments since it also do so in the few-shot learning area.

You can check this at lines 30-33 in utils/return_dataset.py.

meghbhalerao commented 4 years ago

Thanks for letting me know, I saw the lines!

zhaoxin94 commented 4 years ago

@TKKim93 Thanks a lot!