https://github.com/nutszebra/resnet_in_resnet

Implementation of Residual Networks In Residual Networks by chainer (Resnet in Resnet: Generalizing Residual Architectures: https://arxiv.org/abs/1603.08029)
https://github.com/nutszebra/resnet_in_resnet

Last synced: 8 days ago
JSON representation

Implementation of Residual Networks In Residual Networks by chainer (Resnet in Resnet: Generalizing Residual Architectures: https://arxiv.org/abs/1603.08029)

• Host: GitHub
• URL: https://github.com/nutszebra/resnet_in_resnet
• Owner: nutszebra
• License: mit
• Created: 2016-12-17T14:05:54.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2017-01-19T02:20:54.000Z (almost 8 years ago)
• Last Synced: 2024-05-10T09:32:41.836Z (6 months ago)
• Language: Python
• Homepage:
• Size: 111 KB
• Stars: 3
• Watchers: 2
• Forks: 3
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-image-classification - unofficial-chainer : https://github.com/nutszebra/resnet_in_resnet
• awesome-image-classification - unofficial-chainer : https://github.com/nutszebra/resnet_in_resnet

README

        
# What's this

Implementation of Residual Networks In Residual Networks by chainer  

# Dependencies

    git clone https://github.com/nutszebra/resnet_in_resnet.git

    cd resnet_in_resnet

    git submodule init

    git submodule update

# How to run

    python main.py -p ./ -g 0 

# Details about my implementation

All hyperparameters and network architecture are the same as in [[1]][Paper] except for some parts.  

* Data augmentation  

Train: Pictures are randomly resized in the range of [32, 36], then 32x32 patches are extracted randomly and are normalized locally. Horizontal flipping is applied with 0.5 probability.  

Test: Pictures are resized to 32x32, then they are normalized locally. Single image test is used to calculate total accuracy.  

* Learning rate  

Initial learning rate is 0.1. Learning rate is divided by 5 at [150, 225] and I totally run 300 epochs.

# Cifar10 result

| network              | total accuracy (%) |

|:---------------------|-------------------:|

| 18-layer + wide RiR  | 94.99              |

| my implementation    | 94.43               |





# Reference

Resnet in Resnet: Generalizing Residual Architectures [[1]][Paper]

[paper]: https://arxiv.org/abs/1603.08029 "Paper"