[GitHub] thomelane commented on issue #5748: MXNet fcnxs examples: GPU memory usage varies during training.

[GitBox <gi...@apache.org>]

thomelane commented on issue #5748: MXNet fcnxs examples: GPU memory usage varies during training. 
URL: https://github.com/apache/incubator-mxnet/issues/5748#issuecomment-413963845
 
 
   @KeLipeng it looks like you now understand the primary issue of variable gpu memory usage, so going to request for this issue to be closed. Just to confirm, you were seeing this effect because the input images were of different sizes, leading to different sized feature maps and thus different amounts of memory used. 
   
   With regards to the secondary question related to inspecting feature maps. You have to specify it as an output to your model with Module API. What problem does this cause? But with Gluon API (before hybridizing) you can inspect feature maps and kernels really easily. Since you have access to the array values you can calculate summary statistics, plot images, etc. Also I'd recommend [this tutorial](https://mxnet.incubator.apache.org/tutorials/vision/cnn_visualization.html) for visualizing CNNs.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services