[GitHub] lynneChan commented on issue #11137: memory cost problem for different label width when using ssd example

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

lynneChan commented on issue #11137: memory cost problem for different label width when using ssd example
URL: https://github.com/apache/incubator-mxnet/issues/11137#issuecomment-404701765
 
 
   Hi @thomelane, thanks for your replay.
   
    I was using the SSD example code to training a face detector, the WIDER dataset have about 12880 images, so the `batch_size` is fine. And the number of  classes is 2(face or non-face). The code used `DetRecordIter` in `dataset/iterator.py` to get image batches which called another API `mx.io.ImageDetRecordIter`. This API need a `label_width` parameter. 
   
   I keeped all the settings, and just change the `label_width` from 602 to 4784, and the GPU memory cost changes from  6991MB to 10367MB. That's pretty strange, it should not increase so much.

----------------------------------------------------------------
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