pytorch / vision » Build #42

Build Results

TOP 1 ACCURACY TOP 5 ACCURACY
SPEED
MODEL CODE PAPER
ε-REPR
CODE PAPER
ε-REPR
PAPER
AlexNet
(single)
56.1% 57.1% 78.9% -- 376.5
DenseNet-121
74.8% 76.4% 92.2% 93.3% 369.6
DenseNet-161
77.3% -- 93.6% -- 347.9
DenseNet-169
75.9% 77.9% 93.0% 94.1% 361.6
DenseNet-201
77.3% 78.5% 93.5% 94.5% 356.0
Inception V1
70.1% 69.8%
89.7% 89.9% 369.8
Inception V3
69.9% 78.8% 88.9% 94.4% 365.2
MnasNet-A1
73.1% 75.2% 91.3% 92.5% 379.9
MnasNet-A1
(depth multiplier=0.5)
67.2% 68.9% 87.2% -- 377.3
MobileNetV2
71.4% 72.0% 90.1% -- 375.9
ResNet-101
77.3% 78.2% 93.5% 94.0% 358.6
ResNet-152
78.2% 78.6% 94.0% 94.3% 351.1
ResNet-18
69.5% 72.1% 89.1% -- 359.7
ResNet-34 A
73.2% 75.0% 91.3% 92.2% 336.4
ResNet-50
75.9% 77.1% 92.9% 93.3% 372.2
ResNeXt-101 32x8d
79.1% -- 94.4% -- 284.5
ResNeXt-50 32x4d
77.5% -- 93.6% -- 366.6
ShuffleNet V2
(0.5x)
60.0% 60.3% 81.3% -- 384.5
ShuffleNet V2
(1x)
69.0% 69.4% 88.2% -- 384.8
SqueezeNet
57.8% 57.5%
80.3% 80.3% 381.1
SqueezeNet 1.1
57.9% -- 80.4% -- 383.0
VGG-11
68.8% 70.4% 88.6% 89.6% 367.3
VGG-11
(batch-norm)
70.2% 70.4% 89.7% 89.6% 366.2
VGG-13
69.6% 71.3% 89.2% 90.1% 356.2
VGG-13
(batch-norm)
71.4% 71.3% 90.3% 90.1% 362.0
VGG-16
71.4% 74.4% 90.3% 91.9% 356.9
VGG-16
(batch-norm)
73.1% -- 91.4% -- 354.4
VGG-19
72.2% 74.5% 90.7% 92.0% 349.5
VGG-19
(batch-norm)
74.0% -- 91.7% -- 349.5
WRN-101-2-bottleneck
78.6% -- 94.1% -- 323.4
WRN-50-2-bottleneck
78.3% 78.1% 94.0% 94.0% 354.2