log
Classification
ResNet (Residual Network)ResNet (Residual Network)
깊을수록 오히려 학습이 안 되는 Degradation 문제를, 차이만 배우는 Residual Block과 Skip Connection으로 푼 ResNet 정리.Why deeper plain networks got worse (the degradation problem) and how ResNet fixed it by learning only the residual with skip connections.
GoogLeNet (Inception v1)GoogLeNet (Inception v1)
Inception 모듈과 1×1 병목, GAP, 보조 분류기로 깊이는 늘리고 비용은 줄인 GoogLeNet 정리.GoogLeNet went deeper while cutting cost with Inception modules, 1×1 bottlenecks, global average pooling and auxiliary classifiers.
VGGNetVGGNet
3×3 합성곱만 깊게 쌓아 "깊이가 성능을 좌우하는가"에 답한 VGGNet의 아이디어, 한계, PyTorch 활용법.VGGNet answered "does depth matter?" by stacking only 3×3 convolutions — the idea, its limits, and how it is used today in PyTorch.
AlexNetAlexNet
ReLU·GPU·Dropout·데이터 증강으로 거대한 CNN을 실제로 학습시킨 AlexNet 정리와 PyTorch 코드.How AlexNet made a large CNN trainable in practice with ReLU, GPUs, dropout and data augmentation, plus PyTorch code.