ALL-CNN-on-CIFAR10

Pytorch Implementation of ALL-CNN in CIFAR10 Dataset

the model can reach 93% accuracy in CIFAR-10 dataset

Setting in Config

model: select the model to train.

state1-4: the experient in paper using 4 traing stage with different learning-rate.

use_clip: set ture to avoid the gradient explodsion.

data_aug: using crop and flip of images to augment the data.

use_cutout: using cutout to augment the data.

---

Command

use the command below to train the model:

ALL-CNN-C model

python3 main.py train --use_trained_model=False --model='ALL_CNN_C' --lr1=0.1 --lr2=0.05 --lr3=0.01 --lr4=0.001 --weight_decay=0.0001 --use_clip=True --clip=2.0 --use_cutout=False --data_aug=False --class_id=1 --checkpoint_save_name='ALL_CNN_C'

ALL-CNN-C model (with data augmentation)

python3 main.py train --use_trained_model=False --model='ALL_CNN_C' --lr1=0.1 --lr2=0.05 --lr3=0.01 --lr4=0.001 --weight_decay=0.0005 --use_clip=True --clip=2.0 --use_cutout=True --data_aug=True --checkpoint_save_name='ALL_CNN_C_aug'