zhaixingzi

谢谢，我打算用您的代码去复现一下insightface的实验。

您好,训练过程中为什么会出现cos_theta大于1的情况啊?

your code works well, thank you.The method to align face : wuxiang or wenyandong,which is better? Could you give me your advance? Thanks again.

After NCCL is be installed,it works well.But,i'm a little confised: why the original caffe dosen't need to install NCCL to suport multi-gpu? Thank you.

@chichan01 > If I use other way to get 5pts landmarks and use his provide model, result is 98.75 +/-0.194754. could you tell me that which way you used to...

您好，如果我在使用 mxnet native 进行分布式训练的时候，也单独设置scheduler节点， server节点，worker节点，让它们运行在单独的服务器上。这样的话，使用byteps mxnet 与使用 mxnet native 进行训练，效率上会有差异吗？如果有的话，差异主要来自哪里？谢谢。

thank you

@bobzhuyb the number of server is same? between mxnet-native and mxnet-byteps.

你好，我这里还有一个问题。在使用原版mxnet 进行分布式训练的时候，不时的会遇到`Check failed: (my_node_.port) != (-1) bind failed`, 在byteps中，这个问题还存在吗？

@kpzhang93 谢谢。那threshhold里的值越大检测的结果越准确吗？您给的默认值是： > threshold=[0.6 0.7 0.5]