CNTK Examples: Image/Classification/GoogLeNet/BN-Inception

BrainScript

BN-Inception.cntk

The BN-Inception model is implemented according to the model described in Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift.

This implementation achieves 74.938% Top-1 accuracy and 92.346% Top-5 accuracy in our test, which is slightly better than the result in Google’s original paper – 74.8% Top-1 accuracy.

This example with a 256 batch-size should be trained with 8 GPUs.

You could run the example from the current folder using:

mpiexec -n 8 cntk configFile=BN-Inception.cntk

If you would like to run this example with a single card. You could divide the minibatchSize and learningRatesPerMB with a ratio 8 simultaneously.

And run this example using:

cntk configFile=BN-Inception.cntk