This directory contains the code for the paper:
Training CNNs with Selective Allocation of Channels (ICML 2019).
python3
torch >= 0.4.0
torchvision
numpy
tensorboardX
### Train the baseline DenseNet-40 model
$ CUDA_VISIBLE_DEVICES=0 python main.py experiments/cifar10_densenet40.json
### Train DenseNet-40 with channel-selectivity (DenseNet-SConv-40)
$ CUDA_VISIBLE_DEVICES=1 python main.py experiments/cifar10_densenet_sconv40.json
### In case `tensorboard` is installed, you can also track the current training progress
$ tensorboard --logdir=./logs