Open 17011775 opened 5 years ago
VGGNet ( training set : validation set = 10 : 1, Data Augmentation Off, ) batch_size = 3, step_per_epoch = 1500, epochs = 50 : 학습 acc 0.97, 평가 acc 0.53 batch_size = 3, step_per_epoch = 1500, epochs = 35 : 학습 acc 0.97, 평가 acc 0.33 batch_size = 3, step_per_epoch = 1500, epochs = 30 : 학습 acc 0.96, 평가 acc 0.66 batch_size = 3, step_per_epoch = 1500, epochs = 25 : 학습 acc 0.95, 평가 acc 0.33
batch_size = 20, step_per_epoch = 100, epochs = 15 : 학습 acc 0.80, 평가 acc 0.61 batch_size = 20, step_per_epoch = 100 epochs = 20 : 학습 acc 0.88, 평가 acc 0.47
ResNet (training set : validation set = 10 : 1, Data Augmentation Off) batch_size = 20, step_per_epoch = 100, epochs = 50 : 학습 acc 0.98, 평가 acc 0.5
validation accuracy, loss값 변동이 심하고 진전이 잘 안됨-> overfitting에 빠짐 다음번엔 데이터셋 분배 비율을 조정, 데이터 추가 수집, 데이터 어그멘테이션을 적용시킬 예정
VGGNet ( training set : validation set = 10 : 1, Data Augmentation Off, ) batch_size = 3, step_per_epoch = 1500, epochs = 50 : 학습 acc 0.97, 평가 acc 0.53 batch_size = 3, step_per_epoch = 1500, epochs = 35 : 학습 acc 0.97, 평가 acc 0.33 batch_size = 3, step_per_epoch = 1500, epochs = 30 : 학습 acc 0.96, 평가 acc 0.66 batch_size = 3, step_per_epoch = 1500, epochs = 25 : 학습 acc 0.95, 평가 acc 0.33
batch_size = 20, step_per_epoch = 100, epochs = 15 : 학습 acc 0.80, 평가 acc 0.61 batch_size = 20, step_per_epoch = 100 epochs = 20 : 학습 acc 0.88, 평가 acc 0.47
ResNet (training set : validation set = 10 : 1, Data Augmentation Off) batch_size = 20, step_per_epoch = 100, epochs = 50 : 학습 acc 0.98, 평가 acc 0.5
다음번엔 데이터셋 분배 비율을 조정, 데이터 추가 수집, 데이터 어그멘테이션을 적용시킬 예정