DropBlock
Abstract
Deep neural networks often work well when they are over-parameterized and trained with a massive amount of noise and regularization, such as weight decay and dropout. Although dropout is widely used as a regularization technique for fully connected layers, it is often less effective for convolutional layers. This lack of success of dropout for convolutional layers is perhaps due to the fact that activation units in convolutional layers are spatially correlated so information can still flow through convolutional networks despite dropout. Thus a structured form of dropout is needed to regularize convolutional networks. In this paper, we introduce DropBlock, a form of structured dropout, where units in a contiguous region of a feature map are dropped together. We found that applying DropBlock in skip connections in addition to the convolution layers increases the accuracy. Also, gradually increasing number of dropped units during training leads to better accuracy and more robust to hyperparameter choices. Extensive experiments show that DropBlock works better than dropout in regularizing convolutional networks. On ImageNet classification, ResNet-50 architecture with DropBlock achieves 78.13% accuracy, which is more than 1.6% improvement on the baseline. On COCO detection, DropBlock improves Average Precision of RetinaNet from 36.8% to 38.4%.
Installation
Install directly from PyPI:
pip install dropblock
or the bleeding edge version from github:
pip install git+https://github.com/miguelvr/dropblock.git#egg=dropblock
NOTE: Implementation and tests were done in Python 3.6, if you have problems with other versions of python please open an issue.
Usage
For 2D inputs (DropBlock2D):
import torch
from dropblock import DropBlock2D
# (bsize, n_feats, height, width)
x = torch.rand(100, 10, 16, 16)
drop_block = DropBlock2D(block_size=3, drop_prob=0.3)
regularized_x = drop_block(x)
For 3D inputs (DropBlock3D):
import torch
from dropblock import DropBlock3D
# (bsize, n_feats, depth, height, width)
x = torch.rand(100, 10, 16, 16, 16)
drop_block = DropBlock3D(block_size=3, drop_prob=0.3)
regularized_x = drop_block(x)
Scheduled Dropblock:
import torch
from dropblock import DropBlock2D, LinearScheduler
# (bsize, n_feats, depth, height, width)
loader = [torch.rand(20, 10, 16, 16) for _ in range(10)]
drop_block = LinearScheduler(
DropBlock2D(block_size=3, drop_prob=0.),
start_value=0.,
stop_value=0.25,
nr_steps=5
)
probs = []
for x in loader:
drop_block.step()
regularized_x = drop_block(x)
probs.append(drop_block.dropblock.drop_prob)
print(probs)
The drop probabilities will be:
>>> [0. , 0.0625, 0.125 , 0.1875, 0.25, 0.25, 0.25, 0.25, 0.25, 0.25]
The user should include the step() call at the start of the batch loop, or at the the start of a model's forward call.
Check examples/resnet-cifar10.py to see an implementation example.
Implementation details
We use drop_prob instead of keep_prob as a matter of preference, and to keep the argument consistent with pytorch's dropout. Regardless, everything else should work similarly to what is described in the paper.
Benchmark
Reference
[Ghiasi et al., 2018] DropBlock: A regularization method for convolutional networks
TODO
Scheduled DropBlock
Get benchmark numbers
Extend the concept for 3D images