Image Classification on ImageNet LT 2019 (test)
66.2Accuracy (Many)RIDE (3 experts)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RIDE (3 experts)GFlops=4.36 (1.1x), Backbone=ResNet-50, number of experts=32020.10 | 66.2 | 51.7 | 34.9 | 54.9 | |
| RIDE (4 experts)GFlops=5.15 (1.3x), Backbone=ResNet-50, number of experts=42020.10 | 66.2 | 52.3 | 36.5 | 55.4 | |
| RIDE (2 experts)GFlops=3.71 (0.9x), Backbone=ResNet-50, number of experts=22020.10 | 65.8 | 51 | 34.6 | 54.4 | |
| Cross Entropy (CE)GFlops=4.11 (1.0x), Backbone=ResNet-502020.10 | 64 | 33.8 | 5.8 | 41.6 | |
| cRTGFlops=4.11 (1.0x), Backbone=ResNet-502020.10 | 58.8 | 44 | 26.1 | 47.3 | |
| LWSGFlops=4.11 (1.0x), Backbone=ResNet-502020.10 | 57.1 | 45.2 | 29.3 | 47.7 | |
| τ-normGFlops=4.11 (1.0x), Backbone=ResNet-502020.10 | 56.6 | 44.2 | 27.4 | 46.7 | |
| NCMBackbone=ResNet-502020.10 | 53.1 | 42.3 | 26.5 | 44.3 |