Image Classification on ImageNet 1k (val) (Efficiency & Cost Profile)
80Top-1 AccuracyOFALarge
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| OFALargeProgressive Shrinking=true, Fine-tuning epochs=752019.08 | 80 | — | — | 40 | — | 4,200 | 1,200 | 13,000 | — | |
| CondConv-ResNet-50Model Type=CondConv, Backbone=ResNet-50, Number of experts=82019.04 | 78.6 | — | — | — | — | — | — | — | 4,213 | |
| CondConv-EfficientNet-B0Model Type=CondConv, Backbone=EfficientNet-B0, Number of experts=82019.04 | 78.3 | — | — | — | — | — | — | — | 413 | |
| ResNet-50Model Type=Baseline, Backbone=ResNet-502019.04 | 77.7 | — | — | — | — | — | — | — | 4,093 | |
| EfficientNet-B0Model Type=Baseline, Backbone=EfficientNet-B02019.04 | 77.2 | — | — | — | — | — | — | — | 391 | |
| OFAProgressive Shrinking=true, Fine-tuning epochs=752019.08 | 76.9 | — | 58 | 40 | — | 4,200 | 1,200 | 13,000 | — | |
| OFAProgressive Shrinking=true, Fine-tuning epochs=252019.08 | 76.4 | — | 58 | 40 | — | 2,200 | 620 | 6,700 | — | |
| CondConv-MnasNet-A1Model Type=CondConv, Backbone=MnasNet-A1, Number of experts=82019.04 | 76.2 | — | — | — | — | — | — | — | 325 | |
| OFAProgressive Shrinking=true2019.08 | 76 | — | 58 | 40 | 1,200 | 1,200 | 340 | 3,700 | — | |
| MobileNetV3-Large2019.08 | 75.2 | — | 58 | — | — | 7,200 | 1,800 | 22,200 | — | |
| FBNet-C2019.08 | 74.9 | — | — | — | — | 23,000 | 6,500 | 70,400 | — | |
| MnasNet-A1Model Type=Baseline, Backbone=MnasNet-A12019.04 | 74.9 | — | — | — | — | — | — | — | 312 | |
| SinglePathNAS2019.08 | 74.7 | — | — | — | — | 17,000 | 4,800 | 52,000 | — | |
| ProxylessNAS2019.08 | 74.6 | — | 71 | — | — | 20,000 | 5,700 | 61,200 | — | |
| CondConv-MobileNetV2 (1.0x)Model Type=CondConv, Backbone=MobileNetV2, Number of experts=82019.04 | 74.6 | — | — | — | — | — | — | — | 329 | |
| AutoSlim2019.08 | 74.2 | — | 63 | 180 | — | 12,000 | 3,400 | 36,700 | — | |
| NASNet-A2019.08 | 74 | — | — | — | — | — | 544,500 | 5,875,200 | — | |
| MnasNet2019.08 | 74 | — | 70 | — | — | — | 453,800 | 4,896,000 | — | |
| CondConv-MobileNetV1 (1.0x)Model Type=CondConv, Backbone=MobileNetV1, Number of experts=82019.04 | 73.7 | — | — | — | — | — | — | — | 600 | |
| MobileNetV2Training epochs=12002019.08 | 73.5 | — | 66 | 0 | — | 48,000 | 13,600 | 146,900 | — | |
| DARTS2019.08 | 73.1 | — | — | — | — | 14,000 | 4,000 | 42,800 | — | |
| OFAProgressive Shrinking=false2019.08 | 72.4 | — | 59 | 40 | 1,200 | 1,200 | 340 | 3,700 | — | |
| MobileNetV22019.08 | 72 | — | 66 | 0 | — | 6,000 | 1,700 | 18,400 | — | |
| MobileNetV1 (1.0x)Model Type=Baseline, Backbone=MobileNetV12019.04 | 71.9 | — | — | — | — | — | — | — | 567 | |
| MobileNetV2 (1.0x)Model Type=Baseline, Backbone=MobileNetV22019.04 | 71.6 | — | — | — | — | — | — | — | 301 |