js5t3r / pytorch_cutout

A PyTorch implementation of Cutout

PyTorch Implementation of Cutout

Usage

$ python main.py --block_type basic --depth 110 --use_cutout --cutout_size 16 --cutout_prob 1 --outdir results

Results on CIFAR-10

Model	Test Error (median of 5 runs)	Training Time
ResNet-preact-56 w/o Cutout	5.85	98 min
ResNet-preact-56 w/ Cutout	4.96	98 min

w/o Cutout

$ python -u main.py --depth 56 --block_type basic --base_lr 0.2 --seed 7 --outdir results/wo_cutout/00

w/ Cutout

$ python -u main.py --depth 56 --block_type basic --base_lr 0.2 --use_cutout --cutout_size 16 --cutout_prob 1 --seed 7 --outdir results/w_cutout/00

References

• DeVries, Terrance, and Graham W. Taylor. "Improved regularization of convolutional neural networks with cutout." arXiv preprint arXiv:1708.04552 (2017). arXiv:1708.04552, PyTorch implementation