Densely Connected Convolutional Networks

CVPR 2017 Gao HuangZhuang LiuLaurens van der MaatenKilian Q. Weinberger

Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion... (read more)

PDF Abstract

Code


Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT LEADERBOARD
Image Classification CIFAR-10 DenseNet-BC Percentage error 3.46 # 20
Image Classification CIFAR-10 DenseNet Percentage correct 96.54 # 26
Image Classification CIFAR-100 DenseNet Percentage correct 82.62 # 21
Image Classification CIFAR-100 DenseNet-BC Percentage error 17.18 # 12
Image Classification ImageNet DenseNet-169 Top 1 Accuracy 76.2% # 112
Top 5 Accuracy 93.15% # 78
Image Classification ImageNet DenseNet-201 Top 1 Accuracy 77.42% # 99
Top 5 Accuracy 93.66% # 69
Image Classification ImageNet DenseNet-264 Top 1 Accuracy 77.85% # 95
Top 5 Accuracy 93.88% # 67
Image Classification ImageNet DenseNet-121 Top 1 Accuracy 74.98% # 125
Top 5 Accuracy 92.29% # 89
Image Classification SVHN DenseNet Percentage error 1.59 # 10

Results from Other Papers


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK SOURCE PAPER COMPARE
Person Re-Identification DukeMTMC-reID DenseNet-121 [Huang2017Densely] Rank-1 73.16 # 30
MAP 55.08 # 34
Person Re-Identification MSMT17 DenseNet-121 [Huang2017Densely] Rank-1 46.32 # 7
mAP 21.50 # 7
Breast Tumour Classification PCam DenseNet-121 (e) AUC 0.921 # 10
Crowd Counting UCF-QNRF Densenet201 MAE 163 # 3

Methods used in the Paper