George
George
@rcmalli It is a excellent work for VGGFace implement using Keras framework. I have been learned much from your work, thanks a lot.I am new man in deep learning area,...
what's the train dataset you use?@qidiso
what's the hyper-parameters of the first step by using softmax to train the net reaching the accuracy 99.2% on lfw? thank you very much.@qidiso
@EricZgw
好的，明白了，谢谢啊！
@SoftwareGift 按正常情况讲，倒残差模块中膨胀系数6是正常值，该项目中我把t改成1效果也不差。。
我是想裁剪裁剪网络宽度跟深度的， 把block膨胀系数调成1，效果也是很不错的，这个可能跟数据集有关系。
https://github.com/SoftwareGift/FeatherNets_Face-Anti-spoofing-Attack-Detection-Challenge-CVPR2019/issues/42 这个问题我一直没弄明白为什么？@SoftwareGift
what is difference between your model and official model? Did you fix the model to reach the fast results? Could you please show the training method? thank you!@szad670401
change the minSize to 60