Hi, i have a question about the in-domain negative pair generation when calculate proxyploss. In th paper, it says only consider in-domain negative pairs, but in the code (losses.py) i think it also considers cross-domain negative pairs. so only in-domain negative pairs generation isn't implemented in the code, is it right?
Hi, i have a question about the in-domain negative pair generation when calculate proxyploss. In th paper, it says only consider in-domain negative pairs, but in the code (losses.py) i think it also considers cross-domain negative pairs. so only in-domain negative pairs generation isn't implemented in the code, is it right?
Thanks for your attention.