Image Classification on ImageNet Real (val)
91.1Top-1 AccViT-e
Evaluation Results
| Method | Links | |
|---|---|---|
| ViT-e2022.09 | 91.1 | |
| TokenLearner L/8 (24+11)# params.=460M2021.06 | 91.05 | |
| ViT-22BParameters=21.7B, Evaluation Protocol=Linear Probing, Training Data=JFT-3B2023.12 | 90.9 | |
| ViT-G/14# params.=1843M2021.06 | 90.81 | |
| ViT-G2022.09 | 90.8 | |
| TokenLearner L/10 (24+11)# params.=460M2021.06 | 90.75 | |
| ViT-H/14# params.=654M2021.06 | 90.72 | |
| VOLO-D5 + HATParams=295.5M, FLOPs=412G, Test Size=5122022.04 | 90.7 | |
| VOLO-D5Params=295.5M, FLOPs=304G, Test Size=4482022.04 | 90.6 | |
| VOLO-D5 + HATParams=295.5M, FLOPs=304G, Test Size=4482022.04 | 90.6 | |
| VOLO-D5Params=295.5M, FLOPs=412G, Test Size=5122022.04 | 90.6 | |
| BiT-L# params.=928M2021.06 | 90.54 | |
| InternViT-6BParameters=5.9B, Evaluation Protocol=Linear Probing2023.12 | 90.4 | |
| VOLO-D5 + HATParams=295.5M, FLOPs=69.0G, Test Size=2242022.04 | 90.2 | |
| Swin-B uparrow 384 (22k)Params (M)=88, MACs (B)=47.1, Input Resolution=384x3842022.04 | 90 | |
| Mini-DeiT-B uparrow 384Params (M)=44, MACs (B)=56.9, Input Resolution=384x3842022.04 | 89.9 | |
| VOLO-D5Params=295.5M, FLOPs=69.0G, Test Size=2242022.04 | 89.9 | |
| VOLO-D1 + HATParams=26.6M, FLOPs=22.8G, Test Size=3842022.04 | 89.8 | |
| Mini-DeiT-BParams (M)=44, MACs (B)=17.7, Input Resolution=224x2242022.04 | 89.6 | |
| VOLO-D1Params=26.6M, FLOPs=22.8G, Test Size=3842022.04 | 89.6 | |
| DINOv2-gParameters=1.1B, Evaluation Protocol=Linear Probing2023.12 | 89.6 | |
| Mini-Swin-B uparrow 384Params (M)=47, MACs (B)=49.4, Input Resolution=384x3842022.04 | 89.5 | |
| OpenCLIP-GParameters=1.8B, Evaluation Protocol=Linear Probing2023.12 | 89.4 | |
| DeiT-B↑384⚗ / 1000 epochsimage size=384x384, distillation=true, epochs=10002020.12 | 89.3 | |
| EVA-01-CLIP-gParameters=1.1B, Evaluation Protocol=Linear Probing2023.12 | 89.3 | |
| Swin-B (22k)Params (M)=88, MACs (B)=15.4, Input Resolution=224x2242022.04 | 89.2 | |
| VOLO-D1 + HATParams=26.6M, FLOPs=6.8G, Test Size=2242022.04 | 89.2 | |
| DeiT-B↑384⚗image size=384x384, distillation=true2020.12 | 89 | |
| DeiT-B uparrow 384Params (M)=88, MACs (B)=55.7, Input Resolution=384x3842022.04 | 89 | |
| Mini-Swin-BParams (M)=46, MACs (B)=15.7, Input Resolution=224x2242022.04 | 89 | |
| VOLO-D1Params=26.6M, FLOPs=6.8G, Test Size=2242022.04 | 89 | |
| EfficientNet-B6image size=528x5282020.12 | 88.8 | |
| DeiT-B⚗ / 1000 epochsimage size=224x224, distillation=true, epochs=10002020.12 | 88.7 | |
| Mini-Swin-SParams (M)=26, MACs (B)=8.9, Input Resolution=224x2242022.04 | 88.7 | |
| Swin-B uparrow 384Params (M)=88, MACs (B)=47.1, Input Resolution=384x3842022.04 | 88.6 | |
| LiT-g/14Evaluation Protocol=Zero-shot, Backbone=ViT-g/142023.02 | 88.6 | |
| LiT-22BEvaluation Protocol=Zero-shot, Backbone=ViT-22B/142023.02 | 88.6 | |
| Mini-DeiT-SParams (M)=11, MACs (B)=4.7, Input Resolution=224x2242022.04 | 88.4 | |
| LiT-e/14Evaluation Protocol=Zero-shot, Backbone=ViT-e/142023.02 | 88.4 | |
| OpenCLIP-HParameters=0.6B, Evaluation Protocol=Linear Probing2023.12 | 88.4 | |
| EfficientNet-B5image size=456x4562020.12 | 88.3 | |
| DeiT-B⚗image size=224x224, distillation=true2020.12 | 88.3 | |
| EfficientNet-B5Params (M)=30, MACs (B)=9.9, Input Resolution=456x4562022.04 | 88.3 | |
| ViTAEv2-SType=Transformer, Params (M)=19.2, FLOPs (G)=17.8, Input Size=3842022.02 | 88.3 | |
| Swin-B + HATParams=87.8M, FLOPs=15.4G, Test Size=2242022.04 | 88.2 | |
| RegNetY-16GFimage size=224x2242020.12 | 88.1 | |
| RegNetY-16GFParams (M)=84, MACs (B)=15.9, Input Resolution=224x2242022.04 | 88.1 | |
| EfficientNet-B4image size=380x3802020.12 | 88 | |
| EfficientNet-B4Type=CNN, Params (M)=19.3, MACs (G)=8.4, Input Size=3802021.06 | 88 | |
| EfficientNet-B4Type=CNN, Params (M)=19.3, FLOPs (G)=4.2, Input Size=3802022.02 | 88 | |
| ViT-B + HATParams=86.6M, FLOPs=17.6G, Test Size=2242022.04 | 87.9 | |
| Swin-BParams=87.8M, FLOPs=15.4G, Test Size=2242022.04 | 87.9 | |
| DeiT-S⚗ / 1000 epochsimage size=224x224, distillation=true, epochs=10002020.12 | 87.8 | |
| Swin-BParams (M)=88, MACs (B)=15.4, Input Resolution=224x2242022.04 | 87.8 | |
| Swin-SParams=49.6M, FLOPs=8.7G, Test Size=2242022.04 | 87.8 | |
| DeiT-B↑384image size=384x3842020.12 | 87.7 | |
| DeiT-B uparrow 384Params (M)=87, MACs (B)=55.6, Input Resolution=384x3842022.04 | 87.7 | |
| Swin-S + HATParams=49.6M, FLOPs=8.7G, Test Size=2242022.04 | 87.7 | |
| Swin-SParams (M)=50, MACs (B)=8.7, Input Resolution=224x2242022.04 | 87.6 | |
| ViTAEv2-SType=Transformer, Params (M)=19.2, FLOPs (G)=5.7, Input Size=2242022.02 | 87.6 | |
| ViTAE-SType=Transformer, Params (M)=23.6, MACs (G)=20.2, Input Size=3842021.06 | 87.5 | |
| ViTAE-SType=Transformer, Params (M)=23.6, FLOPs (G)=20.2, Input Size=3842022.02 | 87.5 | |
| RegNetY-8GFimage size=224x2242020.12 | 87.4 | |
| RegNetY-8GFType=CNN, Params (M)=39.2, MACs (G)=16, Input Size=2242021.06 | 87.4 | |
| RegNetY-8GFType=CNN, Params (M)=39.2, FLOPs (G)=8, Input Size=2242022.02 | 87.4 | |
| CeiT-SType=Transformer, Params (M)=24.2, MACs (G)=9, Input Size=2242021.06 | 87.3 | |
| Swin-T + HATParams=28.3M, FLOPs=4.5G, Test Size=2242022.04 | 87.3 | |
| CeiT-SType=Transformer, Params (M)=24.2, FLOPs (G)=4.5, Input Size=2242022.02 | 87.3 | |
| Mini-Swin-TParams (M)=12, MACs (B)=4.6, Input Resolution=224x2242022.04 | 87.1 | |
| ViT-BParams=86.6M, FLOPs=17.6G, Test Size=2242022.04 | 87.1 | |
| ViTAE-SType=Transformer, Params (M)=23.6, MACs (G)=5.6, Input Size=2242021.06 | 87 | |
| ViTAE-SType=Transformer, Params (M)=23.6, FLOPs (G)=5.6, Input Size=2242022.02 | 87 | |
| CoE-LargeFLOPs=214M, CondConv=true2021.07 | 86.9 | |
| EfficientNet-B3image size=300x3002020.12 | 86.8 | |
| DeiT-S⚗image size=224x224, distillation=true2020.12 | 86.8 | |
| ViTAE-13MType=Transformer, Params (M)=13.2, MACs (G)=3.4, Input Size=2242021.06 | 86.8 | |
| DeiT-S (distilled)Type=Transformer, Params (M)=22.1, MACs (G)=9.8, Input Size=2242021.06 | 86.8 | |
| T2T-ViT-14Type=Transformer, Params (M)=21.5, MACs (G)=5.2, Input Size=2242021.06 | 86.8 | |
| Swin-TParams=28.3M, FLOPs=4.5G, Test Size=2242022.04 | 86.8 | |
| ViTAE-13MType=Transformer, Params (M)=13.2, FLOPs (G)=3.3, Input Size=2242022.02 | 86.8 | |
| DeiT-S (v2)Type=Transformer, Params (M)=22.1, FLOPs (G)=4.6, Input Size=2242022.02 | 86.8 | |
| T2T-ViT-14Type=Transformer, Params (M)=21.5, FLOPs (G)=5.2, Input Size=2242022.02 | 86.8 | |
| DeiT-Bimage size=224x2242020.12 | 86.7 | |
| CvT-13Type=Transformer, Params (M)=20, MACs (G)=9, Input Size=2242021.06 | 86.7 | |
| DeiT-BParams (M)=86, MACs (B)=17.6, Input Resolution=224x2242022.04 | 86.7 | |
| CvT-13Type=Transformer, Params (M)=20, FLOPs (G)=4.5, Input Size=2242022.02 | 86.7 | |
| Swin-TParams (M)=28, MACs (B)=4.5, Input Resolution=224x2242022.04 | 86.6 | |
| ViT-S + HATParams=22.1M, FLOPs=4.7G, Test Size=2242022.04 | 86.6 | |
| S4L MOAMFLOPs=4B, CondConv=false2021.07 | 86.6 | |
| CoE-LargeFLOPs=194M, CondConv=false2021.07 | 86.5 | |
| RegNetY-4GFimage size=224x2242020.12 | 86.4 | |
| RegNetY-4GFType=CNN, Params (M)=20.6, MACs (G)=8, Input Size=2242021.06 | 86.4 | |
| RegNetY-4GFType=CNN, Params (M)=20.6, FLOPs (G)=4, Input Size=2242022.02 | 86.4 | |
| OFA-595FLOPs=595M, CondConv=false2021.07 | 86 | |
| EfficientNet-B2image size=260x2602020.12 | 85.9 | |
| DeiT-Simage size=224x2242020.12 | 85.7 | |
| DeiT-SType=Transformer, Params (M)=22.1, MACs (G)=9.8, Input Size=2242021.06 | 85.7 | |
| DeiT-SParams (M)=22, MACs (B)=4.6, Input Resolution=224x2242022.04 | 85.7 | |
| ViT-SParams=22.1M, FLOPs=4.7G, Test Size=2242022.04 | 85.7 | |
| DeiT-S (v1)Type=Transformer, Params (M)=22.1, FLOPs (G)=4.6, Input Size=2242022.02 | 85.7 |