Introduction of DenseNet architecture and advantage compared with other CNNs
related with ResNet. They confirm the accuracy of this model by four highly competitive object recognition benchmark tasks.
They said the accuracy of this model is better than any other thing.
high accuracy, parameter & computational efficiency and regularization effect.
Of course. 😁
"collective knowledge" = connection + 새롭게 생성된 feature (쌓아진 느낌)
classifier는 맨 마지막에만 있는게 아니고 implicit deep supervision에도 implicitly 있는 거 아닐까?
deep supervision 설명 : https://eremo2002.tistory.com/116
bottle neck은 왜 Improve computational efficiency?
: https://hoya012.github.io/blog/DenseNet-Tutorial-1/
왜 valid set으로만 했을까???