Weiyang Liu

@BeginnerW Large datasets such as MS-1M have a lot of overlapping labels with LFW, so you should not directly train on MS and test on LFW. FYI, if you train...

@hardegg Original softmax loss trainied on MS-1M could also easily give you very high accuracy. For A-softmax, you can consider three things: 1) use longer iteration 2) try smaller lambda...

SphereFace can work well with VGGface2. Previously, we have successfully trained SphereFace on VGG2. Besides that, the results of [InsightFace](https://github.com/deepinsight/insightface) also has validated this. Your testing procedure may somehow be...

@yxchng 如果我没有理解错，你是用casia训练的模型，然后直接用vgg2的数据去测试，而protocol就是用的你描述的这个。我没检查你的测试代码（假设你的预处理、测试等部分没问题），但是我觉得你还是可以做个sanity check，你可以拿除了sphereface之类的模型，比如center loss，比如普通的softmax loss，做一个同样的测试，看看结果怎么样。我觉得在casia上训练的模型（比如我们提供的sphereface pretrained model），在vgg2上不能达到你说的那种“类内最大距离小于类间最小距离”，这个也说不定是有可能的，毕竟sphereface只是要求在训练集上尽可能产生大的angular margin，这个在如果完全不同的测试集上，能不能完全泛化过去，确实是不好说的。

We obtained 99.3% accuracy for the pretrained model in our experiment. Our reported result is not a mistake. The reason that you only get 99.27% may vary. Usually, it is...

I am not quite sure what causes the 0.03% accuracy drop in your case. The reason can vary. But you can retrain the model from scratch using our code. It...

@zeakey @yxchng First of all, I knew many people have reproduced the 99.3% accuracy using the pretrained model. You guys may do something wrong in the pipeline. It may be...

@zeakey For evaluating our SphereFace-20, we used cuda8 w/o cudnn. The version may or may not be the reason. To be honest, I am not so sure. I think the...

The training accuracy in the training protext is not the true accuracy of the model. Because when you are training the network with A-Softmax (say with m=4), you are actually...

It should be easy to process a video using SphereFace. You just do recognition frame by frame. We do not provide a single script that produces the video. It is...