VGG16

1. Dataset

• Train dataset: Cifar 10 dataset with 2 augmentation

2. Trainning

• Using VGG 16
• learning rate 0.01 for 1~95 epoch
• learning rate 0.001 for 96 ~ 110 epoch
• learning rate 0.0001 for 110~ 150 epoch

3. Result

• Best model at 115 epoch
• Test acc: 96.67%      Test Loss: 0.13332
• Val acc: 93.06%    Val Loss: 0.2888

4. Test

python3 test.py -i <IMAGE_PATH>

Other models' acc

1. 92.64% (https://github.com/kuangliu/pytorch-cifar)
2. over 90% (https://www.kaggle.com/xhlulu/vgg-16-on-cifar10-with-keras-beginner)