D-MMD
D-MMD copied to clipboard
Published
20 hours ago •
djidje
djidje
Results
2
D-MMD issues
Sort by
recently updated
recently updated
newest added
Thanks for your work and releasing your code . after reading your code , I have a question . Did you regard images of same person as a tracklet in...
我仔细的阅读了一下您公开的代码，发现你在源域和目标域上使用的都是RandomIdentitySampler，所以您在计算MMD loss时直接将一个instances作为同一身份处理，而论文中提到的是将同一tracklet中的帧作为WC样本，将来自不同tracklet的帧作为BC样本，是否有问题？期待您的解惑
