This is my second time to upload code to github, if there is something wrong with README and LICENSE, please let me know. I will modify or delete the code or project if necessary.
other implementations of DropBlock:
An implementation of DropBlock
in PyTorch. DropBlock
is a replacement for dropout which zero-es
entire spatial blocks instead of single pixels. For more information, check the paper:
DropBlock: A regularization method for convolutional networks
Inspired by DHZS:tf-dropblock, I modified the code from Randl:DropBlock-pytorch, and use max pooling instead of conv to get the mask of DropBlock(named DropBlock2DV2 in the code). All the code except DropBlock2DV2 comes from Randl:DropBlock-pytorch, you can go to Randl:DropBlock-pytorch and (原)DropBlock A regularization method for convolutional networks for more information.