Without triplet loss the performance is quite limited?

[kilickaya]

Hello! Thanks for the nice work.

In my experiments solely using LogRatioLoss leads to much inferior results for pose recognition. When I combine it with binary triplet loss then I can get something more reasonable. Is it expected? Have you combined triplet loss with LogRatioLoss in your experiments within the paper?

Thanks!