Image Classification on ImageNet-1k 1.0 (val) (Top-1 Accuracy and Hardware Latency)
79.4Top-1 AccMobileOne-S4
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MobileOne-S4Architecture Type=Convolutional, FLOPs (M)=2978, Params (M)=14.8, Resolution=224x224, Distillation=false2022.06 | 79.4 | 26.6 | 0.95 | 1.86 | |
| MobileViT-SArchitecture Type=Transformer, FLOPs (M)=1792, Params (M)=5.6, Resolution=224x224, Distillation=false2022.06 | 78.4 | 30.76 | — | 9.21 | |
| RepVGG-B1Architecture Type=Convolutional, FLOPs (M)=11800, Params (M)=51.8, Resolution=224x224, Distillation=false2022.06 | 78.4 | 193.7 | 3.17 | 3.73 | |
| StarNet-S4Params (M)=7.5, FLOPs (M)=1075, Latency benchmarking batch size=12024.03 | 78.4 | 9.4 | 3.7 | 1 | |
| MobileOne-S3Architecture Type=Convolutional, FLOPs (M)=1896, Params (M)=10.1, Resolution=224x224, Distillation=false2022.06 | 78.1 | 16.47 | 0.76 | 1.53 | |
| EdgeViT-XSParams (M)=6.8, FLOPs (M)=1166, Latency benchmarking batch size=12024.03 | 77.5 | 18.3 | 12.1 | 3.5 | |
| MobileOne-S2Architecture Type=Convolutional, FLOPs (M)=1299, Params (M)=7.8, Resolution=224x224, Distillation=false2022.06 | 77.4 | 14.87 | 0.72 | 1.18 | |
| MobileOne-S2Params (M)=7.8, FLOPs (M)=1299, Latency benchmarking batch size=12024.03 | 77.4 | 8.9 | 2 | 1 | |
| StarNet-S3Params (M)=5.8, FLOPs (M)=757, Latency benchmarking batch size=12024.03 | 77.3 | 6.7 | 2.7 | 0.9 | |
| EfficientNet-B0Architecture Type=Convolutional, FLOPs (M)=390, Params (M)=5.3, Resolution=224x224, Distillation=false2022.06 | 77.1 | 28.71 | 1.35 | 1.72 | |
| EfficientNet-B0Params (M)=5.3, FLOPs (M)=390, Latency benchmarking batch size=12024.03 | 77.1 | 8.8 | 3.4 | 1.6 | |
| RepVGG-A2Architecture Type=Convolutional, FLOPs (M)=5100, Params (M)=25.5, Resolution=224x224, Distillation=false2022.06 | 76.5 | 93.43 | 2.41 | 2.41 | |
| FasterNet-T1Params (M)=7.6, FLOPs (M)=855, Latency benchmarking batch size=12024.03 | 76.2 | 9.7 | 3.3 | 0.9 | |
| MobileNeXt-x1.4Architecture Type=Convolutional, FLOPs (M)=590, Params (M)=6.1, Resolution=224x224, Distillation=false2022.06 | 76.1 | 18.06 | 1.04 | 1.27 | |
| MobileOne-S1Architecture Type=Convolutional, FLOPs (M)=825, Params (M)=4.8, Resolution=224x224, Distillation=false2022.06 | 75.9 | 13.04 | 0.66 | 0.89 | |
| MobileOne-S1Params (M)=4.8, FLOPs (M)=825, Latency benchmarking batch size=12024.03 | 75.9 | 6 | 1.5 | 0.9 | |
| MixNet-SArchitecture Type=Convolutional, FLOPs (M)=256, Params (M)=4.1, Resolution=224x224, Distillation=false2022.06 | 75.8 | 40.09 | 2.41 | 1.13 | |
| GhostNet1.3Params (M)=7.3, FLOPs (M)=226, Latency benchmarking batch size=12024.03 | 75.7 | 11 | 3.9 | 9.7 | |
| MobileNetV3-LArchitecture Type=Convolutional, FLOPs (M)=219, Params (M)=5.4, Resolution=224x224, Distillation=false2022.06 | 75.2 | 17.09 | 3.8 | 1.09 | |
| MNASNet-A1Architecture Type=Convolutional, FLOPs (M)=312, Params (M)=3.9, Resolution=224x224, Distillation=false2022.06 | 75.2 | 24.06 | 0.95 | 1 | |
| MobileV3-LParams (M)=5.4, FLOPs (M)=219, Latency benchmarking batch size=12024.03 | 75.2 | 5.2 | 2.5 | 11.4 | |
| RepVGG-B0Architecture Type=Convolutional, FLOPs (M)=3100, Params (M)=14.3, Resolution=224x224, Distillation=false2022.06 | 75.1 | 55.97 | 1.45 | 1.82 | |
| ShuffleNetV2-2.0Architecture Type=Convolutional, FLOPs (M)=591, Params (M)=7.4, Resolution=224x224, Distillation=false2022.06 | 74.9 | 20.85 | 4.76 | 1.08 | |
| ShuffleV2-2.0Params (M)=7.4, FLOPs (M)=591, Latency benchmarking batch size=12024.03 | 74.9 | 9.7 | 2.6 | 19.9 | |
| MobileViT-XSArchitecture Type=Transformer, FLOPs (M)=941, Params (M)=2.3, Resolution=224x224, Distillation=false2022.06 | 74.8 | 27.21 | — | 6.97 | |
| StarNet-S2Params (M)=3.7, FLOPs (M)=547, Latency benchmarking batch size=12024.03 | 74.8 | 4.5 | 2 | 0.7 | |
| MobileNetV2-x1.4Architecture Type=Convolutional, FLOPs (M)=585, Params (M)=6.9, Resolution=224x224, Distillation=false2022.06 | 74.7 | 15.67 | 0.8 | 1.36 | |
| MobileV2-1.4Params (M)=6.9, FLOPs (M)=585, Latency benchmarking batch size=12024.03 | 74.7 | 5.4 | 2.8 | 1.1 | |
| RepVGG-A1Architecture Type=Convolutional, FLOPs (M)=2400, Params (M)=12.8, Resolution=224x224, Distillation=false2022.06 | 74.5 | 47.15 | 1.42 | 1.68 | |
| EdgeViT-XXSParams (M)=4.1, FLOPs (M)=559, Latency benchmarking batch size=12024.03 | 74.4 | 12.6 | 8.9 | 1.8 | |
| MobileNeXt-x1.0Architecture Type=Convolutional, FLOPs (M)=311, Params (M)=3.4, Resolution=224x224, Distillation=false2022.06 | 74 | 16.04 | 1.02 | 0.92 | |
| GhostNet1.0Params (M)=5.2, FLOPs (M)=141, Latency benchmarking batch size=12024.03 | 73.9 | 7 | 3.6 | 7.9 | |
| StarNet-S1Params (M)=2.9, FLOPs (M)=425, Latency benchmarking batch size=12024.03 | 73.5 | 4.3 | 2.3 | 0.7 | |
| MobileV3-L0.75Params (M)=4, FLOPs (M)=155, Latency benchmarking batch size=12024.03 | 73.3 | 4.4 | 2.2 | 10.9 | |
| ConViT-tinyArchitecture Type=Transformer, FLOPs (M)=1000, Params (M)=5.7, Resolution=224x224, Distillation=false2022.06 | 73.1 | 28.95 | — | 10.99 | |
| Mobileformer-96Architecture Type=Transformer, FLOPs (M)=96, Params (M)=4.6, Resolution=224x224, Distillation=false2022.06 | 72.8 | 37.36 | — | 16.95 | |
| ShuffleV2 1.5Params (M)=3.5, FLOPs (M)=299, Latency benchmarking batch size=12024.03 | 72.6 | 4.9 | 2.2 | 5.9 | |
| RepVGG-A0Architecture Type=Convolutional, FLOPs (M)=1400, Params (M)=8.3, Resolution=224x224, Distillation=false2022.06 | 72.4 | 43.61 | 1.23 | 1.28 | |
| DeiT-tinyArchitecture Type=Transformer, FLOPs (M)=1300, Params (M)=5.9, Resolution=224x224, Distillation=false2022.06 | 72.2 | 16.68 | 1.78 | 4.78 | |
| MobileNetV2-x1.0Architecture Type=Convolutional, FLOPs (M)=300, Params (M)=3.4, Resolution=224x224, Distillation=false2022.06 | 72 | 13.65 | 0.69 | 0.98 | |
| MobileV2-1.0Params (M)=3.4, FLOPs (M)=300, Latency benchmarking batch size=12024.03 | 72 | 3.2 | 2.2 | 0.9 | |
| FasterNet-T0Params (M)=3.9, FLOPs (M)=338, Latency benchmarking batch size=12024.03 | 71.9 | 5.7 | 2.5 | 0.7 | |
| MobileOne-S0Architecture Type=Convolutional, FLOPs (M)=275, Params (M)=2.1, Resolution=224x224, Distillation=false2022.06 | 71.4 | 10.55 | 0.56 | 0.79 | |
| MobileOne-S0Params (M)=2.1, FLOPs (M)=275, Latency benchmarking batch size=12024.03 | 71.4 | 2.2 | 1.1 | 0.7 | |
| PiT-tiArchitecture Type=Transformer, FLOPs (M)=710, Params (M)=4.9, Resolution=224x224, Distillation=false2022.06 | 71.3 | 16.37 | 1.97 | 8.81 | |
| MobileNetV1Architecture Type=Convolutional, FLOPs (M)=575, Params (M)=4.2, Resolution=224x224, Distillation=false2022.06 | 70.6 | 10.65 | 0.58 | 0.95 | |
| ShuffleNetV2-1.0Architecture Type=Convolutional, FLOPs (M)=146, Params (M)=2.3, Resolution=224x224, Distillation=false2022.06 | 69.4 | 16.6 | 4.58 | 0.68 | |
| ShuffleV2-1.0Params (M)=2.3, FLOPs (M)=146, Latency benchmarking batch size=12024.03 | 69.4 | 3.8 | 2.2 | 4.1 | |
| MobileViT-XXSArchitecture Type=Transformer, FLOPs (M)=373, Params (M)=1.3, Resolution=224x224, Distillation=false2022.06 | 69 | 23.03 | — | 4.7 | |
| Mobileformer-52Architecture Type=Transformer, FLOPs (M)=52, Params (M)=3.6, Resolution=224x224, Distillation=false2022.06 | 68.7 | 29.23 | — | 9.02 | |
| Mobileformer-52Params (M)=3.6, FLOPs (M)=52, Latency benchmarking batch size=12024.03 | 68.7 | 26 | 8.3 | 6.6 | |
| MobileNetV3-SArchitecture Type=Convolutional, FLOPs (M)=56, Params (M)=2.5, Resolution=224x224, Distillation=false2022.06 | 67.4 | 10.38 | 3.74 | 0.83 | |
| MobileV3-SParams (M)=2.9, FLOPs (M)=66, Latency benchmarking batch size=12024.03 | 67.4 | 2.6 | 1.8 | 6.5 | |
| GhostNet0.5Params (M)=2.6, FLOPs (M)=42, Latency benchmarking batch size=12024.03 | 66.2 | 4.8 | 2.9 | 10 | |
| MobileV3-S0.75Params (M)=2.4, FLOPs (M)=44, Latency benchmarking batch size=12024.03 | 65.4 | 2.5 | 1.8 | 5.5 |