FeatherNets_Face-Anti-spoofing-Attack-Detection-Challenge-CVPR2019 icon indicating copy to clipboard operation
FeatherNets_Face-Anti-spoofing-Attack-Detection-Challenge-CVPR2019 copied to clipboard
verified publisher icon SoftwareGift

过拟合问题

Open LiangXiaoEr opened this issue 6 years ago • 3 comments

你好，我按照原代码的设置，用FeatherNetB在CASIA_SURF的原始数据集的深度图上进行训练和验证的时候发现训练集的accuracy和loss曲线是正常的，验证集上的accuracy是先升后降，loss是先降后升的，但是best模型的ACER值又能达到论文上面的精度，请问这个问题该怎么处理？

LiangXiaoEr avatar Sep 23 '19 03:09 LiangXiaoEr

选择性能最好的

SoftwareGift avatar Sep 23 '19 14:09 SoftwareGift

@LiangXiaoEr 您好　您能把数据分享一下吗

S130111 avatar Nov 12 '19 06:11 S130111

@S130111 你签个协议书发给CASIA_SURF的作者就行了

  1. Download, read the Contest Rules, and sign the agreement,link
  2. Send the your signed agreements to: Jun Wan, [email protected]

LiangXiaoEr avatar Nov 12 '19 07:11 LiangXiaoEr