Skip to content
Implementation of GoogLeNet by chainer (Going Deeper with Convolutions:
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
trainer @ 4a84c45 submodule Dec 4, 2016
.gitignore updated Dec 9, 2016
.gitmodules submodule Dec 4, 2016
accuracy.jpg result Dec 9, 2016 fixed model definition Dec 4, 2016
loss.jpg result Dec 9, 2016 fixed bug Dec 4, 2016 optimizer Dec 4, 2016

What's this

Implementation of GoogLeNet by chainer


git clone
cd googlenet
git submodule init
git submodule update

How to run

python -p ./ -g 0 

Details about my implementation

  • Data augmentation
    Train: Pictures are randomly resized in the range of [256, 512], then 224x224 patches are extracted randomly and are normalized locally. Horizontal flipping is applied with 0.5 probability.
    Test: Pictures are resized to 384x384, then they are normalized locally. Single image test is used to calculate total accuracy.

  • Auxiliary classifiers
    No implementation

  • Learning rate
    As [[1]][Paper] said, learning rate are multiplied by 0.96 at every 8 epochs. The description about initial learning rate can't be found in [[1]][Paper], so initial learning is setted as 0.0015 that is found in [[2]][Paper2].

  • Weight decay
    The description about weight decay can't be found in [[1]][Paper], so by using [[2]][Paper2] and [[3]][Paper3] I guessed that weight decay is 2.0*10^-4.

Cifar10 result

network depth total accuracy (%)
my implementation 22 91.33


total accuracy


Going Deeper with Convolutions [[1]][Paper]
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift [[2]][Paper2]
Rethinking the Inception Architecture for Computer Vision [[3]][Paper3]
[paper]: "Paper" [paper2]: "Paper2" [paper3]: "Paper3"

You can’t perform that action at this time.