You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@singa.apache.org by GitBox <gi...@apache.org> on 2021/11/24 08:49:14 UTC
[GitHub] [singa] lzjpaul opened a new issue #904: Move data augmentation outside of training iterations for better training efficiency
lzjpaul opened a new issue #904:
URL: https://github.com/apache/singa/issues/904
In examples/largedataset_cnn/train_largedata.py, the augmentation should be moved outside of training iterations.
For benchmark datasets, e.g., CIFAR-10 and ImageNet, the augmentation method is pre-defined before the model training, so the augmentation does not need to be carried out on the fly in each iteration.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: dev-unsubscribe@singa.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [singa] lzjpaul closed issue #904: Move data augmentation outside of training iterations for better training efficiency
Posted by GitBox <gi...@apache.org>.
lzjpaul closed issue #904:
URL: https://github.com/apache/singa/issues/904
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: dev-unsubscribe@singa.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [singa] nudles commented on issue #904: Move data augmentation outside of training iterations for better training efficiency
Posted by GitBox <gi...@apache.org>.
nudles commented on issue #904:
URL: https://github.com/apache/singa/issues/904#issuecomment-978715786
But, the data augmentation is done with some randomness, e.g., random cropping offset and random rotation degrees. If you move it outside of the training iterations, you may lose this randomness. As a result, you may not get much benefit from data augmentation.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: dev-unsubscribe@singa.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [singa] lzjpaul commented on issue #904: Move data augmentation outside of training iterations for better training efficiency
Posted by GitBox <gi...@apache.org>.
lzjpaul commented on issue #904:
URL: https://github.com/apache/singa/issues/904#issuecomment-983347926
The data augmentation is remained inside each training iteration for better classification accuracy
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: dev-unsubscribe@singa.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org