Image Classification on ImageNet Real 1k (val)
90.2Top-1 AccCaiT-M36↑448Y
Evaluation Results
| Method | Links | |
|---|---|---|
| CaiT-M36↑448YParameters=271M, FLOPs=247.8B, Training Resolution=224, Test Resolution=4482021.03 | 90.2 | |
| CaiT-M48↑448YParameters=356M, FLOPs=329.6B, Training Resolution=224, Test Resolution=4482021.03 | 90.2 | |
| Refined-ViT-L↑448Params=81M, FLOPS=98.0B, Train size=224, Test size=4482021.06 | 90.1 | |
| Fix-EfficientNet-B8Parameters=87M, FLOPs=89.5B, Training Resolution=672, Test Resolution=8002021.03 | 90 | |
| CaiT-M36↑384YParameters=271M, FLOPs=173.3B, Training Resolution=224, Test Resolution=3842021.03 | 90 | |
| Fix-EfficientNet-B8Params=87M, FLOPS=89.5B, Train size=672, Test size=8002021.06 | 90 | |
| NFNet-F6+SAMParameters=438M, FLOPs=377.3B, Training Resolution=448, Test Resolution=5762021.03 | 89.9 | |
| CaiT-S36↑384YParameters=68M, FLOPs=48.0B, Training Resolution=224, Test Resolution=3842021.03 | 89.8 | |
| CaiT-S36↑384Params=68M, FLOPS=48.0B, Train size=224, Test size=3842021.06 | 89.8 | |
| Refined-ViT-L↑384Params=81M, FLOPS=69.1B, Train size=224, Test size=3842021.06 | 89.7 | |
| CaiT-S48↑384Parameters=89M, FLOPs=63.8B, Training Resolution=224, Test Resolution=3842021.03 | 89.5 | |
| EfficientNet-B7 AdvPropParameters=66M, FLOPs=37.0B, Training Resolution=600, Test Resolution=6002021.03 | 89.4 | |
| NFNet-F3Parameters=255M, FLOPs=114.8B, Training Resolution=320, Test Resolution=4162021.03 | 89.4 | |
| NFNet-F4Parameters=316M, FLOPs=215.3B, Training Resolution=384, Test Resolution=5122021.03 | 89.4 | |
| LionModel=ViT-L/16, Resolution=384, Optimizer=Lion, RandAug + Mixup=false, Training Protocol=Pre-train on ImageNet-21K then fine-tune on ImageNet2023.02 | 89.35 | |
| DeiT-B↑384Y 1000 epochsParameters=87M, FLOPs=55.5B, Training Resolution=224, Test Resolution=3842021.03 | 89.3 | |
| CaiT-M36Params=271M, FLOPS=53.7B, Train size=224, Test size=2242021.06 | 89.3 | |
| Refined-ViT-M↑384Params=55M, FLOPS=49.2B, Train size=224, Test size=3842021.06 | 89.3 | |
| NFNet-F5Parameters=377M, FLOPs=289.8B, Training Resolution=416, Test Resolution=5442021.03 | 89.2 | |
| CaiT-S36↑384Parameters=68M, FLOPs=48.0B, Training Resolution=224, Test Resolution=3842021.03 | 89.2 | |
| Refined-ViT-LParams=81M, FLOPS=19.1B, Train size=224, Test size=2242021.06 | 89.1 | |
| NFNet-F1Parameters=133M, FLOPs=35.5B, Training Resolution=224, Test Resolution=3202021.03 | 88.9 | |
| NFNet-F2Parameters=194M, FLOPs=62.6B, Training Resolution=256, Test Resolution=3522021.03 | 88.9 | |
| CaiT-S36YParameters=68M, FLOPs=13.9B, Training Resolution=224, Test Resolution=2242021.03 | 88.9 | |
| NFNet-F1Params=133M, FLOPS=35.5B, Train size=224, Test size=3202021.06 | 88.9 | |
| Refined-ViT-S↑384Params=25M, FLOPS=24.5B, Train size=224, Test size=3842021.06 | 88.9 | |
| Refined-ViT-MParams=55M, FLOPS=13.5B, Train size=224, Test size=2242021.06 | 88.9 | |
| LionModel=ViT-B/16, Resolution=384, Optimizer=Lion, RandAug + Mixup=false, Training Protocol=Pre-train on ImageNet-21K then fine-tune on ImageNet2023.02 | 88.84 | |
| AdamWModel=ViT-L/16, Resolution=384, Optimizer=AdamW, RandAug + Mixup=false, Training Protocol=Pre-train on ImageNet-21K then fine-tune on ImageNet2023.02 | 88.78 | |
| AdamWModel=ViT-B/16, Resolution=384, Optimizer=AdamW, RandAug + Mixup=false, Training Protocol=Pre-train on ImageNet-21K then fine-tune on ImageNet2023.02 | 88.61 | |
| LV-ViT-MParams=56M, FLOPS=16.0B, Train size=224, Test size=2242021.06 | 88.4 | |
| EfficientNet-B5Parameters=30M, FLOPs=9.9B, Training Resolution=456, Test Resolution=4562021.03 | 88.3 | |
| EfficientNet-B5Params=30M, FLOPS=9.9B, Train size=456, Test size=4562021.06 | 88.3 | |
| Refined-ViT-SParams=25M, FLOPS=7.2B, Train size=224, Test size=2242021.06 | 88.3 | |
| RegNetY-16GFParameters=84M, FLOPs=16.0B, Training Resolution=224, Test Resolution=2242021.03 | 88.1 | |
| NFNet-F0Parameters=72M, FLOPs=12.4B, Training Resolution=192, Test Resolution=2562021.03 | 88.1 | |
| NFNet-F0Params=72M, FLOPS=12.4B, Train size=192, Test size=2562021.06 | 88.1 | |
| LV-ViT-SParams=26M, FLOPS=6.6B, Train size=224, Test size=2242021.06 | 88.1 | |
| CaiT-S36Parameters=68M, FLOPs=13.9B, Training Resolution=224, Test Resolution=2242021.03 | 88 | |
| DeiT-B↑384Parameters=86M, FLOPs=55.4B, Training Resolution=224, Test Resolution=3842021.03 | 87.7 | |
| DeiT-B↑384Params=86M, FLOPS=55.4B, Train size=224, Test size=3842021.06 | 87.7 | |
| DeiT-BParameters=86M, FLOPs=17.5B, Training Resolution=224, Test Resolution=2242021.03 | 86.7 | |
| DeiT-BParams=86M, FLOPS=17.5B, Train size=224, Test size=2242021.06 | 86.7 | |
| LionModel=ViT-B/16, Optimizer=Lion, RandAug + Mixup=true, Training Protocol=Train from scratch on ImageNet2023.02 | 86.15 | |
| DeiT-SParameters=22M, FLOPs=4.6B, Training Resolution=224, Test Resolution=2242021.03 | 85.7 | |
| DeiT-SParams=22M, FLOPS=4.6B, Train size=224, Test size=2242021.06 | 85.7 | |
| AdamWModel=ViT-B/16, Optimizer=AdamW, RandAug + Mixup=true, Training Protocol=Train from scratch on ImageNet2023.02 | 85.46 | |
| LionModel=ViT-S/16, Optimizer=Lion, RandAug + Mixup=true, Training Protocol=Train from scratch on ImageNet2023.02 | 85.25 | |
| AdamWModel=ViT-S/16, Optimizer=AdamW, RandAug + Mixup=true, Training Protocol=Train from scratch on ImageNet2023.02 | 84.61 | |
| ViT-B/16Parameters=86M, FLOPs=55.4B, Training Resolution=224, Test Resolution=3842021.03 | 83.6 | |
| ViT-B/16Params=86M, FLOPS=55.4B, Train size=224, Test size=3842021.06 | 83.6 | |
| AdamWModel=ResNet-50, Optimizer=AdamW, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 82.72 | |
| LionModel=ResNet-50, Optimizer=Lion, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 82.72 | |
| LionModel=ViT-S/16, Optimizer=Lion, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 82.64 | |
| LionModel=ViT-B/16, Optimizer=Lion, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 82.57 | |
| SGDModel=ResNet-50, Optimizer=SGD, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 82.39 | |
| ViT-L/16Parameters=307M, FLOPs=190.7B, Training Resolution=224, Test Resolution=3842021.03 | 82.2 | |
| ViT-L/16Params=307M, FLOPS=190.7B, Train size=224, Test size=3842021.06 | 82.2 | |
| AdamWModel=ViT-S/16, Optimizer=AdamW, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 81.94 | |
| AdamWModel=ViT-B/16, Optimizer=AdamW, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 80.64 | |
| LionModel=Mixer-B/16, Optimizer=Lion, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 76.6 | |
| LionModel=Mixer-S/16, Optimizer=Lion, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 76.19 | |
| AdamWModel=Mixer-S/16, Optimizer=AdamW, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 75.71 | |
| AdamWModel=Mixer-B/16, Optimizer=AdamW, RandAug + Mixup=false, Training Protocol=Train from scratch on ImageNet2023.02 | 73.92 |