cifar-10作为分类的基础数据集,已经接近探索极限,目前Gpipe, EfficientNet-b7等网络,已经获得了接近99%的验证精度
事实上,很多较大的网络,都能很轻易地将训练精度达到100%
因此,以cifar-10作为AI contest的目标数据集,吐槽之余,应该确立一个有意义的目标:
以某个精度较高的基础网络作为baseline,试验各种去过拟合或提高泛化性的算法,最大化压缩验证精度与训练精度的gap
-
Notifications
You must be signed in to change notification settings - Fork 0
xingyueye/Pytorch_cifar
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
Prepare for SRCX AI Contest
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published