Implement the code from the Paper, all the architecture and parameters are the same as paper.

The results are shown as below:

Accuracy and loss of Inception_v4 with 100 classes in ILSVRC2012:

Accuracy and loss of Inception_ResNet_v1 with 100 classes in ILSVRC2012:

Accuracy and loss of Inception_ResNet_v2 with 1000 classes in ILSVRC2012: