CIFAR-100-CS543

How to ace on a custom CIFAR-100 dataset (CS543 UIUC)

WRNs:

• Wide Residual Networks

My implementation in wrn.py

AutoAugment:

• AutoAugment: Learning Augmentation Policies from Data

In autoaugment.py, there are 25 subpolicies trained from a reduced CIFAR-10 dataset, we use those subpolicies in our CIFAR-100 training.

Implementation ported from https://github.com/DeepVoltaire/AutoAugment

Cutout:

• Improved Regularization of Convolutional Neural Networks with Cutout

My implementation in cutout.py

RandomErasing:

• Random Erasing Data Augmentation

random_erasing.py contains the original author's implementation https://github.com/zhunzhong07/Random-Erasing

Implementation Details

Default script will run WRN-28-10: wide residual network with depth=28 and k=10 (widen factor)
Batch size: 128

Learning rate settings:

epoch	learning rate	weight decay	Optimizer	Momentum	Nesterov
0 ~ 60	0.1	0.0005	Momentum	0.9	true
61 ~ 120	0.02	0.0005	Momentum	0.9	true
121 ~ 160	0.004	0.0005	Momentum	0.9	true
161 ~ 200	0.0008	0.0005	Momentum	0.9	true