How to achieve dynamic training?

Klitter / Pyramidal_Person_ReID

A simple reproduction of Pyramidal Person Re-IDentification via Multi-Loss Dynamic Training (CVPR2019).

MIT License

34 stars 16 forks source link

How to achieve dynamic training? #5

Open swg209 opened 4 years ago

swg209 commented 4 years ago

@Klitter Hi, I am trying to achieve the dynamic training, but I meet some problem. As the paper says, each iteration choose one sample method in ID-balanced hard triplet and randomized. So I want to make sure these question: Is the iteration means epoch? Is there any achievement of the mutil-loss dynamic Training code for reference?

tomFoxxxx commented 3 years ago

是的。epoch。第二个问题没看懂 achievement？打错字了吧老哥

furoce commented 3 years ago

他的意思是不是：现有有没有达到论文中多个loss动态训练效果的训练代码？