Closed kksgit closed 4 years ago
OK, I'll answer it by myself. This function is for multi-GPU training. For more details, if someone has questions, u can google 'multigpu training'.
By the way, for Chinese users, I recommend you the blog as following. https://www.cnblogs.com/jins-note/p/9674515.html
Hi Charles,
Good job on PointNet! I am confused about the tf_util.py. U use the _variable_on_cpu() to get variable by CPU. Does it mean that the weights and bias are stored on CPU cache? What is the reason you did that? Save GPU memory? Or?
Rhett