Skip to content
Code for the paper "Training CNNs with Selective Allocation of Channels" (ICML 2019)
Branch: master
Clone or download
Latest commit 8aafd93 May 14, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
experiments
models
.gitignore
LICENSE
README.md
datasets.py
main.py
selective_convolution.py
utils.py

README.md

Introduction

This directory contains the code for the paper:

Training CNNs with Selective Allocation of Channels (ICML 2019).

Requirements

  • python3
  • torch >= 0.4.0
  • torchvision
  • numpy
  • tensorboardX

How to run

### 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
You can’t perform that action at this time.