An interesting question.

uta-smile / TVT

Code of TVT: Transferable Vision Transformer for Unsupervised Domain Adaptation, WACV 2023

MIT License

65 stars 11 forks source link

An interesting question. #16

Open 1a1a1a1a1a1a1a1 opened 4 months ago

1a1a1a1a1a1a1a1 commented 4 months ago

Hi author, I ported your code to my domain and after training for a while, I noticed that my loss in the source domain kept changing during training, but the accuracy in the target domain surprisingly stayed exactly the same. Have you ever encountered this situation on your side?

hong-god commented 3 months ago

I'm in the same situation as you.