fiveai / lame Goto Github PK

Hello! I tried to reproduce the results of your paper in Fig. 4 for the baseline method and the LAME method, but I always get slightly different results compared to the output on paper. I have tried different batch sizes (16.64) but most of the problems are in the I.I.D. with Likelihood Shift + Prior Shift scenario. I also shared the results of each experiment with different batch sizes. Could you suggest how can I solve this problem? Because I want to try your approach for my experiments and I need to reproduce the results like in the post.

Thanks in advance!

Hi,Interesting work! I'd like to inquire about the relation of Figure 3 w.r.t TTA. Generalisation differs a lot than TTA what message do you try to deliver with tuning on domain i and evaluating on j this seems as domain generalisation not TTA as TTA should evaluate on the same test sample from the same domain after adaptation. Clarification would be appreciated!

Hi, authors, it is an amazing method. Can this method apply to the regression task? Could you please share any opinions on it? Many thanks in advance!