Skip to content

A collection of codes for 'how far can we go with MNIST' challenge

Notifications You must be signed in to change notification settings

hwalsuklee/how-far-can-we-go-with-MNIST

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 

Repository files navigation

How far can we go with MNIST??

A collection of implementations for 'how far can we go with MNIST' challenge, which has been held in TF-KR at April 2017.

List of Implementations

Kyung Mo Kweon

Junbum Cha

  • Test error : 0.24%
  • Features : tensorflow, ensemble of 3 models (VGG-like with batch size 64/128, resnet 32layers), best accuracy with a single model is 99.74%, data augmentation (rotation, shift, zoom)
  • https://github.com/khanrc/mnist

Jehoon Shin

Owen Song

Kiru Park

  • Test error : 0.30%
  • Features : tflearn, ensemble of 11 models (5 conv-nets, 3 highway-nets, 3 rnn), weights for ensemble are also trained, data augmentation (shift, rotation, blur)
  • https://github.com/kirumang/mnist_kr

Mintae Kim

Juyoung Lee

  • Test error : 0.37%
  • Features : tensorflow, a single model (conv3-conv3-conv3-pool-conv5-conv-conv5-conv5-conv7-conv7-fc-fc-fc-fc), data augmentation (elastic transform)
  • https://github.com/uptown/TF-Mnist

Hyungchan Kim

Taekang Woo

Hc Chae

  • Test error : 0.46%
  • Features : tensorflow, ensemble of 5 models obtained with same hyper-params and same architecture (VGG-like), best accuracy with a single model is 0.9935, data augmentation (scale, rotation)
  • https://github.com/chaeso/dnn-study

Junhyun Lee

Sungsub Woo

  • Test error : 0.48%
  • Features : keras, ensemble of 50 models obtained with same hyper-params and same architecture (3 conv-layers, 1 fc-layer), data augmentation (infmnist)
  • https://github.com/sungchi/mnist/

Byeongki Jeong

Sungho Park

Wonseok Jeon

Byungsun Bae

Hyun Seok Jeong

Sung Kim

Acknowledgements