Implementation of Densely Connected Convolutional Networks (DCCN) by chainer
git clone https://github.com/nutszebra/dense_net.git
cd dense_net
git submodule init
git submodule update
python main.py -p ./ -e 300 -b 64 -g 0 -s 1 -trb 4 -teb 4 -lr 0.1
All hyperparameters and network architecture are the same as in [1] except for data-augmentation.
- 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.
As a result, I could confirm 95.12% total accuracy at epoch 250 and this result has almost the same accuracy as reported by [1].
network | depth | k | total accuracy (%) |
---|---|---|---|
DCCN [1] | 40 | 12 | 94.76 |
my implementation | 40 | 12 | 95.12 |
DCCN [1] | 100 | 12 | 95.9 |
DCCN [1] | 100 | 24 | 96.26 |