[GitHub] Ishitori commented on issue #12063: MxNet Pre-trained Prediction Doc Not Working

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

Ishitori commented on issue #12063: MxNet Pre-trained Prediction Doc Not Working
URL: https://github.com/apache/incubator-mxnet/issues/12063#issuecomment-420743508
 
 
   @prathik-naidu I think that model may actually have been trained without normalization. 
   
   I receive worse results if I apply normalization to the data. The top one is: `probability=0.060943, class=n01930112 nematode, nematode worm, roundworm` which is clearly worse than original one: `probability=0.249607, class=n02119022 red fox, Vulpes vulpes`
   
   If I change model to Gluon one (`resnet18 = vision.resnet18_v1(pretrained=True)`), then it is clear that it was trained with normalization in place. With normalization I get: `probability=8.849780, class=n02124075 Egyptian cat`, without I get: `probability=60.358135, class=n04270147 spatula`
   
   So, I am not sure how this model was trained. It could be that it actually didn't use normalization. What results do you get if you apply normalization?

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