A resnet 34-layer model for MS-Celeb-1M challeng 1
Due to the large amont of training images (~8.5 millions), we reduce the total training epochs to 30 epochs, and lower the learning rate from 0.1 to 0.01 after 10 epochs, and to 0.001 after 20 epochs. All other settings are the same with original ResNet training setting.
You can download the trained model from Google Drive
tensorflow, tensorpack