Skip to content

vgg

Jump to bottom
chenyaofo edited this page May 15, 2019 · 1 revision

Model Details for VGG

VGG for cifar is not mentioned in the original paper, Very Deep Convolutional Networks for Large-Scale Image Recognition. We rewrite its imagenet version, torchvision/models/vgg.py, to adapt to cifar dataset with minimal modification.

The modifications include:

  • remove the AdaptiveAvgPool2d layer
  • change the number of features in linear layers to 512->512->512->num_classes, while original one is 512*7*7->4096->4096->num_classes
Clone this wiki locally