CD1_OpenCode

2022년 1학기 컴퓨터공학과 캡스톤디자인1 오픈코드팀

Multiple Data Augmentation(CutMix+Mixup)

Experimental results

Model + Method	Train Top1 Accuracy	Validation Top1 Accuracy	Test Top1 Accuracy
ResNet18	100	65.41	65.46
ResNet18+CutMix	98.62	73.57	73.15
ResNet18+Mixup	97.38	73.53	73.32
ResNet18+CutMixup	99.98	67.11	67.45
ResNet18+Multiple Data Augmentation(CutMix+Mixup)	92.80	75.12	74.95
ResNet34	100	66.30	65.50
ResNet34+CutMix	98.14	73.03	74.54
ResNet34+Mixup	97.15	72.92	72.96
ResNet34+CutMixup	99.95	68.04	67.78
ResNet34+Multiple Data Augmentation(CutMix+Mixup)	92.39	73.65	73.95
DenseNet169	100	71.59	71.45
DenseNet169+CutMix	98.33	77.57	77.72
DenseNet169+Mixup	76.19	76.45	77.07
DenseNet169+CutMixup	98.89	76.33	76.87
DenseNet169+Multiple Data Augmentation(CutMix+Mixup)	73.74	77.14	77.95

Environment

• Python 3.7.13
• Pytorch 1.11.0
• CUDA 11.3
• Ubuntu 18.04.5 LTS
• Tesla P100