Skip to content

Latest commit

 

History

History

densenet

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
README.md
README.md
 
 
densenet.py
densenet.py
 
 
results.png
results.png
 
 

README.md

This is a sample implementation of DenseNet-40 applied to CIFAR-10, see the Densely Connected Convolutional Network paper for details. The original paper implementation using Torch can be found on github.

This uses the python version of the CIFAR-10 dataset. Most implementation details should be similar to the reference paper/implementation.

  • The SGD optimizer is used with a momentum of 0.9. The initial learning rate is 0.1, it gets divided by 10 at epochs 150 and 225.
  • Weight decay is set to 1e-4.
  • Batch size is 64.
  • This implements DenseNet-40, i.e. 3 dense blocks of 12 layers each. The growth rate is set to 12.
  • A dropout with a keep probability of 0.8 is used after each convolution. Convolutions don't use any bias.

The resulting accuracy is 93% on CIFAR-10 without data augmentation which is similar to what is reported in the paper.

Results