Image Classification on CIFAR-10 (Top-1 Error)
0.0162Top-1 Error (%)Divide and Co-training (WRN-40-10)
Evaluation Results
| Method | Links | |
|---|---|---|
| Divide and Co-training (WRN-40-10)S (number of small networks)=4, epochs=1800, MParams=55.9, GFLOPs=8.122020.11 | 0.0162 | |
| Divide and Co-training (WRN-28-10)S (number of small networks)=4, epochs=1800, MParams=36.5, GFLOPs=5.282020.11 | 0.0168 | |
| Divide and Co-training (Shake-Shake 26 2×96d)S (number of small networks)=4, epochs=300, MParams=26.3, GFLOPs=3.822020.11 | 0.0169 | |
| PyramidNet-272MParams=26.2, GFLOPs=4.552020.11 | 0.017 | |
| Divide and Co-training (Shake-Shake 26 2×96d)S (number of small networks)=2, epochs=300, MParams=23.3, GFLOPs=3.382020.11 | 0.0175 | |
| Divide and Co-training (WRN-28-10)S (number of small networks)=2, epochs=1800, MParams=35.8, GFLOPs=5.162020.11 | 0.0181 | |
| Divide and Co-training (ResNeXt-29, 8×64d)S (number of small networks)=2, epochs=1800, MParams=35.1, GFLOPs=5.492020.11 | 0.0194 | |
| Shake-Shake 26 2×96d w/ AutoAugmentMParams=26.2, GFLOPs=3.782020.11 | 0.02 | |
| Shake-Shake 26 2×96d w/ RandAugmentMParams=26.2, GFLOPs=3.782020.11 | 0.02 | |
| Shake-Shake 26 2×96d (re-implementation)epochs=300, MParams=26.2, GFLOPs=3.782020.11 | 0.02 | |
| Shake-Shake 26 2×96d (re-implementation)epochs=1800, MParams=26.2, GFLOPs=3.782020.11 | 0.02 | |
| Divide and Co-training (WRN-28-10)S (number of small networks)=4, epochs=300, MParams=36.5, GFLOPs=5.282020.11 | 0.0201 | |
| Divide and Co-training (WRN-40-10)S (number of small networks)=4, epochs=300, MParams=55.9, GFLOPs=8.122020.11 | 0.0201 | |
| Divide and Co-training (WRN-28-10)S (number of small networks)=2, epochs=300, MParams=35.8, GFLOPs=5.162020.11 | 0.0206 | |
| Divide and Co-training (ResNeXt-29, 8×64d)S (number of small networks)=2, epochs=300, MParams=35.1, GFLOPs=5.492020.11 | 0.0212 | |
| SE-ResNet-164 (re-implementation)epochs=1800, MParams=2.49, GFLOPs=0.262020.11 | 0.0219 | |
| WRN-40-10 (re-implementation)epochs=1800, MParams=55.8, GFLOPs=8.082020.11 | 0.0219 | |
| Divide and Co-training (SE-ResNet-164)S (number of small networks)=2, epochs=1800, MParams=2.81, GFLOPs=0.292020.11 | 0.022 | |
| ResNeXt-29, 8×64d (re-implementation)epochs=1800, MParams=34.4, GFLOPs=5.402020.11 | 0.0223 | |
| WRN-28-10 (re-implementation)epochs=300, MParams=36.5, GFLOPs=5.252020.11 | 0.0228 | |
| WRN-40-10 w/ Cutout, Mixup, Fixup-initMParams=55.9, GFLOPs=8.082020.11 | 0.023 | |
| WRN-40-10 (re-implementation)epochs=300, MParams=55.8, GFLOPs=8.082020.11 | 0.0233 | |
| WRN-28-10 (re-implementation)epochs=1800, MParams=36.5, GFLOPs=5.252020.11 | 0.0241 | |
| WRN-28-10 w/ AutoAugmentMParams=36.5, GFLOPs=5.252020.11 | 0.0268 | |
| ResNeXt-29, 8×64d (re-implementation)epochs=300, MParams=34.4, GFLOPs=5.402020.11 | 0.0269 | |
| WRN-28-10 w/ RandAugmentMParams=36.5, GFLOPs=5.252020.11 | 0.027 | |
| Divide and Co-training (SE-ResNet-164)S (number of small networks)=2, epochs=300, MParams=2.81, GFLOPs=0.292020.11 | 0.0284 | |
| SE-ResNet-164 (re-implementation)epochs=300, MParams=2.49, GFLOPs=0.262020.11 | 0.0298 | |
| ResNeXt-29, 8×64dMParams=34.5, GFLOPs=5.412020.11 | 0.0365 | |
| WRN-28-10MParams=36.5, GFLOPs=5.252020.11 | 0.04 | |
| SE-ResNet-164MParams=2.51, GFLOPs=0.262020.11 | 0.0439 | |
| ResNet-164MParams=1.73, GFLOPs=0.252020.11 | 0.0546 |