List your company
ResearchFormerly WizWand

Image Classification on MNIST rotated (test)

0.59Test Error (%)Sim(2) Separable G-CNN
-6.146439.324384.795130.2657Apr 14, 2014Jul 16, 2015Oct 17, 2016Jan 18, 2018Apr 22, 2019Jul 23, 2020Oct 25, 2021
Updated 8mo ago

Evaluation results

MethodLinks
Sim(2) Separable G-CNNSeparable configuration=along dilation and rotations dimensions, Train-time augmentation=continuous rotations2021.100.59
SE(2) Separable G-CNNTrain-time augmentation=continuous rotations2021.100.66
Sim(2) Separable G-CNNSeparable configuration=along dilation and rotations dimensions2021.100.66
E(2)-NNTrain-time augmentation=continuous rotations2021.100.68
E(2)-equivariant steerable CNNsgroup=D16 -> C16, representation=regular2019.110.682
E(2)-equivariant steerable CNNsgroup=C16, representation=quotient2019.110.705
PDO-eConvData Augmentation=Continuous rotations, Parameters=0.65M, Layers=72020.070.709
SFCNNData Augmentation=Continuous rotations2020.070.714
[7]group=C16, representation=regular2019.110.714
SFCNNTrain-time augmentation=continuous rotations2021.100.714
E2CNNData Augmentation=Continuous rotations2020.070.716
E(2)-equivariant steerable CNNsgroup=C16, representation=regular2019.110.716
SFCNN2021.100.88
PTN-CNNData Augmentation=Continuous rotations2020.070.89
SE(2) Separable G-CNN2021.100.89
RotEqNettest-time augmentation=true2016.121.01
RotEqNet2016.121.09
RotEqNetData Augmentation=Continuous rotations2020.071.09
[13]group=C17, representation=regular/vector2019.111.09
ORN-8(ORAlign)time(s)=17.8, params(%)=31.412017.011.12
TI-pooling2016.121.2
[40]2019.111.2
ORN-8(ORPooling)time(s)=17.9, params(%)=12.872017.011.21
LieConv2021.101.24
TIPooling(x8)time(s)=126.7, params(%)=100.002017.011.26
ORN-4(ORPooling)time(s)=8, params(%)=7.952017.011.33
ORN-8(None)time(s)=17.5, params(%)=31.412017.011.33
ORN-4(ORAlign)time(s)=8.1, params(%)=15.912017.011.34
ORN-8(ORPooling)time(s)=17.9, params(%)=12.872017.011.37
RED-NN2021.101.39
ORN-8(ORAlign)time(s)=17.8, params(%)=31.412017.011.42
ORN2016.121.54
OR-TIPooling(with augmentation)Augmentation=true2017.011.54
OR-TIPoolingData Augmentation=Continuous rotations2020.071.54
ORN-4(None)time(s)=7.9, params(%)=15.912017.011.55
ORN-8(None)time(s)=17.5, params(%)=31.412017.011.57
H-Net2016.121.69
ORN-4(ORAlign)time(s)=8.1, params(%)=15.912017.011.69
H-NetData Augmentation=Continuous rotations2020.071.69
[12]group=SO(2), representation=irreducible2019.111.69
H-Net2021.101.69
STN(affine)time(s)=18.5, params(%)=100.402017.011.82
ORN-4(ORPooling)time(s)=8, params(%)=7.952017.011.84
PDO-eConvparams=26k, Data Augmentation=None2020.071.87
ORN-4(None)time(s)=7.9, params(%)=15.912017.011.88
TIPooling(with augmentation)Augmentation=true2017.011.93
STN(rotation)time(s)=18.7, params(%)=100.392017.011.93
RotEqNetmodel_variant=only scalar field2016.122.01
Baseline CNNtime(s)=16.4, params(%)=100.002017.012.19
TI-Poolingparams=13.3M, Data Augmentation=None2020.072.2
ORN-8(ORAlign)Number of orientations=8, Alignment=ORAlign2017.012.25
ORN-8 (ORNAlign)params=0.53M, Data Augmentation=None2020.072.25
P4CNNgroup=p4, pooling=no pooling over rotations in intermediate layers2016.022.28
G-CNNparams=25k, Data Augmentation=None2020.072.28
[6]group=C4, representation=regular2019.112.28
G-CNN2021.102.28
STN(affine)time(s)=18.5, params(%)=100.402017.012.52
Baseline CNNtime(s)=16.4, params(%)=100.002017.012.82
STN(rotation)time(s)=18.7, params(%)=100.392017.012.88
P4CNNRotationPoolinggroup=p4, pooling=coset max-pooling over rotations2016.023.21
[6]group=C4, representation=regular/scalar2019.113.21
Schmidt & Roth (2012)2016.023.98
Sohn & Lee (2012)2016.024.2
TIRBM2016.124.2
TIRBM2014.044.2
TIRBM2017.014.2
TIRBMData Augmentation=None2020.074.2
CNN2017.014.34
Z2CNNlayers=7 layers of 3x3 convolutions, channels=20 per layer, group=Z22016.025.03
CNNparams=22k, Data Augmentation=None2020.075.03
PCANet-2Number of stages=22014.047.37
PCANet-22017.017.37
PCANet-2Data Augmentation=None2020.077.37
ScatNet-22014.047.48
ScatNet-22017.017.48
ScatNet-2Data Augmentation=None2020.077.48
LDANet-2Number of stages=22014.047.52
PCANet-1Number of stages=1, k1=132014.048.3
RandNet-2Number of stages=22014.048.47
CAE-22014.049.66
Larochelle et al. (2007)2016.0210.38
PCANet-1Number of stages=12014.0410.55
LDANet-1Number of stages=12014.0411.4
RandNet-1Number of stages=12014.0414.25
ORN-8(ORAlign)time(s)=17.8, params(%)=31.412017.0116.21
ORN-8(ORPooling)time(s)=17.9, params(%)=12.872017.0116.67
ORN-4(ORPooling)time(s)=8, params(%)=7.952017.0127.74
ORN-4(ORAlign)time(s)=8.1, params(%)=15.912017.0127.92
STN(rotation)time(s)=18.7, params(%)=100.392017.0155.59
Baseline CNNtime(s)=16.4, params(%)=100.002017.0156.28
STN(affine)time(s)=18.5, params(%)=100.402017.0156.44
ORN-8(None)time(s)=17.5, params(%)=31.412017.0158.98
ORN-4(None)time(s)=7.9, params(%)=15.912017.0159.67
SFCNNsInitialization=CoeffInit, Data Augmentation=train time augmentation2017.1171.4
SFCNNsInitialization=CoeffInit2017.1188
SFCNNsInitialization=HeInit2017.1195.7
Marcos et al.Data Augmentation=test time augmentation2017.11101
Marcos et al.2017.11109
Laptev et al.2017.11120
Worrall et al.2017.11169