/Imagenet

Multi-GPU to train Imagenet by different models

Primary LanguagePython

Imagenet

Multi-GPU to train Imagenet by different models

First download the Imagenet dataset using the official script and transform to TFRecord

like

1 2

Train the model using "inference.py"

Here we use 4 gpu cards to train 1000K steps, and the training summary is like this

3 4 5

The training learning rate is show like this

6

Test the model

Here we use a single gpu and single image crop the top5 accuracy is 0.9048

7