List your company
ResearchFormerly WizWand

Classification on SVHN (test)

1.12Error RateColornet-40-48
1.05521.49261.932.3674Feb 18, 2013Feb 15, 2014Feb 12, 2015Feb 10, 2016Feb 6, 2017Feb 3, 2018Feb 1, 2019
Updated 4mo ago

Evaluation results

MethodLinks
Colornet-40-48No of Parameters=19.0M2019.021.12——
DeVries & TaylorNormalization=Yes2019.011.3——
BatchNorm + Mixup + CutoutNormalization=Yes2019.011.4——
Fixup-init + Mixup + CutoutNormalization=No2019.011.4——
Zagoruyko & KomodakisNormalization=Yes2019.011.5——
WRN-16-8dropout=true2016.051.54——
CoPaNet-RDropout=true, Depth=164, Params=15.7 M, Pathways (k)=2, Width (m)=32017.091.58——
RoR-3-WRN58-4+SDParameters=13.3M2016.081.59——
DenseNetDropout=false, Depth=100, Params=27.2 M, Growth Rate=242017.091.59——
DenseNetGrowth Rate (k)=24, Depth=100, Params=27.2M2016.081.59——
Colornet-40-12No of Parameters=1.75M2019.021.59——
Mixed max-avg poolingPooling Parameters=1 per layer/channel/region2015.091.64——
WRN28-10Parameters=36.5M2016.081.64——
Wide ResNet with DropoutDepth=16, Params=2.7M2016.081.64——
SimpNet2018.021.648——
DenseNetGrowth Rate (k)=12, Depth=100, Params=7.0M2016.081.67——
CRMN-28Parameters=>40M2016.081.68——
Tree+Max-Avg poolingPooling Parameters=1 per pool layer2015.091.69——
Gen. Pool2016.031.69——
Lee et al.Normalization=No2019.011.7——
CoPaNet-RDropout=true, Depth=164, Params=7.00 M, Pathways (k)=2, Width (m)=22017.091.72——
CoPaNetDropout=true, Depth=164, Params=27.9 M, Pathways (k)=2, Width (m)=42017.091.73——
DenseNet-BCGrowth Rate (k)=24, Depth=250, Params=15.3M2016.081.74——
Densenet-BC-250-24No of Parameters=15.3M2019.021.74——
ResNet with stochastic depth2016.031.75——
ResNets-110+SDParameters=1.7M, Note=152-layer2016.081.75——
Stochastic DepthDropout=false, Depth=110, Params=1.7 M2017.091.75——
ResNet with Stochastic DepthDepth=110, Params=1.7M2016.081.75——
ResNet with Stochastic Depth2018.021.75——
Resnet-101No of Parameters=1.7M2019.021.75——
DenseNet-BCGrowth Rate (k)=12, Depth=100, Params=0.8M2016.081.76——
Densenet-BC-100-12No of Parameters=0.8M2019.021.76——
Recurrent CNN (R-CNN)2015.091.77——
RCNN-1922015.111.77——
R-CNN2016.031.77——
Scalable BO2016.031.77——
RCNN-192channels=1922015.111.77——
DenseNetGrowth Rate (k)=12, Depth=40, Params=1.0M2016.081.79——
HybridNetNb. labeled images=1000, Backbone=ResNet2018.071.8——
ResNet with constant depth2016.031.8——
MINk (number of maxout pieces)=52015.111.81——
Maxout Network In NetworkDropout=true, Params=1.6 M, Pathways (k)=52017.091.81——
CoPaNetDropout=true, Depth=164, Params=6.98 M, Pathways (k)=2, Width (m)=22017.091.83——
HybridNetNb. labeled images=500, Backbone=ResNet2018.071.85——
CoPaNetDropout=true, Depth=164, Params=1.75 M, Pathways (k)=2, Width (m)=12017.091.86——
FractalNetParameters=30M2016.081.87——
FractalNet with Dropout/Drop-pathDepth=21, Params=38.6M2016.081.87——
Fractalnet with dropoutNo of Parameters=38.6M, dropout=true2019.021.87——
Our baseline2015.091.91——
State-of-the-art2015.031.92——
Deeply-Supervised Nets (DSN)2015.091.92——
DSN2015.111.92——
Deeply Supervised2016.031.92——
DSN2016.081.92——
Deeply Supervised NetDepth=-, Params=-2016.081.92——
Deeply Supervised Net2018.021.92——
Deeply-supervised nets2015.111.92——
5 × CONV. NET + DROPCONNECTData Augmentation=true2013.121.93——
DropConnect2013.121.94——
DropConnect2015.091.94——
Dropconnectdata augmentation=true, multiple model voting=true2015.111.94——
DropConnect2016.031.94——
Drop-connect2015.111.94——
Human performance2015.112——
FractalNetDepth=21, Params=38.6M2016.082.01——
ResNet (reported by [13])Depth=110, Params=1.7M2016.082.01——
ResNetreported by=[66] (2016)2018.022.01——
FractalnetNo of Parameters=38.6M2019.022.01——
Mean Teacher ResNetNb. labeled images=1000, Backbone=ResNet, reproduced_by_authors=true2018.072.05——
Multi-digit Number Recognition2013.122.16——
Mean Teacher ResNetNb. labeled images=500, Backbone=ResNet, reproduced_by_authors=true2018.072.33——
NIN + Dropout2013.122.35——
Network in Network (NiN)2015.092.35——
NIN2015.112.35——
Net in Net2016.032.35——
NIN2016.082.35——
Network In NetworkDropout=true, Params=0.98 M2017.092.35——
Network in NetworkDepth=-, Params=-2016.082.35——
Network in Network2018.022.35——
Network in Network2015.112.35——
N-in-N2019.022.35——
CONV. NET + PROBOUTData Augmentation=false2013.122.39——
Probabilistic Maxout2015.092.39——
FitNet2015.092.42——
FitNet2016.082.42——
Conv. maxoutdropout=true2013.022.47——
Conv. maxout + Dropout2013.122.47——
CONV. NET + MAXOUTData Augmentation=false2013.122.47——
Maxout Networks2015.092.47——
Maxout networkk (number of maxout pieces)=22015.112.47——
Maxout2016.032.47——
Maxout NetworkDropout=true, Pathways (k)=22017.092.47——
Conv. Maxout+DropoutMaxout=true, Dropout=true2015.112.47——
Pi modelNumber of labels=73257, Implementation=Replicated/This paper baseline, Architecture=13-layer ConvNet2017.032.5——
Mean TeacherNumber of labels=73257, Architecture=13-layer ConvNet2017.032.5——
Pi modelNumber of labels=73257, Implementation=Reported in [13], Architecture=13-layer ConvNet2017.032.54——
Rectifiersdropout=true, synthetic translation=true2013.022.68——
Rectifier + Dropout + Synthetic Translationdata augmentation=synthetic translation2013.122.68——
CONV. NET + DROPOUTData Augmentation=true2013.122.68——
Temporal EnsemblingNumber of labels=73257, Architecture=13-layer ConvNet2017.032.74——