You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2017/10/25 01:22:31 UTC

[GitHub] qingzhouzhen opened a new pull request #8417: pvanet for classfication on Imagenet

qingzhouzhen opened a new pull request #8417: pvanet for classfication on Imagenet
URL: https://github.com/apache/incubator-mxnet/pull/8417
 
 
   article adress : [Pvanet:Deep but Lightweight Neural Neural Networks for Real-time Object Detection](https://arxiv.org/abs/1608.08021)
   pvanet for classification on ImageNet, the accuracy of the article is 70%, I checked the shape of parameter to make sure the net is same with the author implement with Caffe, but I still got a accuracy of 66.28%, I trained with:
     intialize?mx.init.Normal()
     batch-size=512
     lr-factor=0.316
     lr=0.1
     num_epochs=160
     lr_step_epochs='30,60,80,100,120,140'
   
   result is :
   
     INFO:root:Epoch[158] Batch [2350]       Speed: 624.06 samples/sec       accuracy=0.696992       cross-entropy=1.213310  top_k_accuracy_5=0.890508
     INFO:root:Epoch[158] Batch [2400]       Speed: 624.90 samples/sec       accuracy=0.698438       cross-entropy=1.205256  top_k_accuracy_5=0.894531
     INFO:root:Epoch[158] Batch [2450]       Speed: 625.89 samples/sec       accuracy=0.690547       cross-entropy=1.232207  top_k_accuracy_5=0.890508
     INFO:root:Epoch[158] Batch [2500]       Speed: 627.79 samples/sec       accuracy=0.695352       cross-entropy=1.215836  top_k_accuracy_5=0.892695
     INFO:root:Epoch[158] Train-accuracy=0.683594
     INFO:root:Epoch[158] Train-cross-entropy=1.284250
     INFO:root:Epoch[158] Train-top_k_accuracy_5=0.888672
     INFO:root:Epoch[158] Time cost=2048.858
     INFO:root:Saved checkpoint to "pvanet-models/pvanet-0159.params"
     INFO:root:Epoch[158] Validation-accuracy=0.662847
     INFO:root:Epoch[158] Validation-cross-entropy=1.374140 
     INFO:root:Epoch[158] Validation-top_k_accuracy_5=0.869958
   
   can this be accepted by MXNet?

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