Why your test accuracy is lower than 90% with vanilla Resnet50? about mixup HOT 3 CLOSED

The symbol of resnet50 is writen by mxnet https://github.com/apache/incubator-mxnet/tree/master/example/image-classification/symbols, there have many versions. And i havenot do any
optimizion for it. All the results are based on this baseline.

from mixup.

I reread the Resnet paper http://cn.arxiv.org/pdf/1512.03385.pdf.
On cifar 10,ResNet-44 gets 7.17% error, ResNet-56 gets 6.97% error in this paper.
So, I think resnet-50 baseline should be close to 93%.
If your baseline is too low with the claimed accuracy, I think the comparision is not confident.

from mixup.

i will have a try on other resnet version in mxnet office example. thx for your suggestion.

from mixup.