Densely Connected Convolutional Network implementation by Chainer

Implementation by Chainer. Original paper is Densely Connected Convolutional Network.

Requirements

• python 2.7

• Chainer 1.19.0+ (Neural network framework)

Start training

For example, run,

python train.py --gpus 0 --batchsize 64 --dataset cifar10 --lr 0.1 --depth 100 --growth_rate 24 --split_size 4

Show possible options

python train.py --help

Sample results

• Cifar-10 (batchsize=64, depth=100, growth_rate=24, with data augmentation)

Original paper reported 3.74% validation error under the same conditions.

• Cifar-100 (batchsize=64, depth=100, growth_rate=24, with data augmentation)

Original paper reported 19.25% validation error under the same conditions.