search

google / sg2im

Code for "Image Generation from Scene Graphs", Johnson et al, CVPR 2018
Apache License 2.0
1.29k stars 231 forks source link

Switching to eval mode after 100k iterations #14

Closed amirbar closed 5 years ago

amirbar commented 5 years ago

Hi @jcjohnson,

Thanks for sharing the code and for the great work on this.

While looking at the training code, I've came across the following line: https://github.com/google/sg2im/blob/master/scripts/train.py#L510

It seems in this line you switch the model to eval mode (although existence of batchnorm/dropout layers) and create a new optimizer instance.

I'm wondering what is the justifications for this or whether you have found this to be useful for particular reason?

Thanks you so much for your time.

Best, Amir