Image Classification on ImageNet-1K 1.0 (val) (Top-1 Accuracy and MACs)
86.4Top-1 AccuracyCAFormer-B36
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CAFormer-B36MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=99, Resolution=3842022.10 | 86.4 | 72.2 | — | |
| MViTv2-LMetaFormer=Yes, Mixing Type=Attn, Params (M)=218, Resolution=3842022.10 | 86.3 | 140.2 | — | |
| MaxViT-BMetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=120, Resolution=3842022.10 | 86.3 | 74.2 | — | |
| CAFormer-M36MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=56, Resolution=3842022.10 | 86.2 | 42 | — | |
| iFormer-LMetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=87, Resolution=3842022.10 | 85.8 | 45.3 | — | |
| CoAtNet-3MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=168, Resolution=3842022.10 | 85.8 | 107.4 | — | |
| CAFormer-S36MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=39, Resolution=3842022.10 | 85.7 | 26 | — | |
| CoAtNet-2MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=75, Resolution=3842022.10 | 85.7 | 49.8 | — | |
| ConvFormer-B36MetaFormer=Yes, Mixing Type=Conv, Params (M)=100, Resolution=3842022.10 | 85.7 | 66.5 | — | |
| ConvFormer-M36MetaFormer=Yes, Mixing Type=Conv, Params (M)=57, Resolution=3842022.10 | 85.6 | 37.7 | — | |
| MViTv2-BMetaFormer=Yes, Mixing Type=Attn, Params (M)=52, Resolution=3842022.10 | 85.6 | 36.7 | — | |
| SLaK-BMetaFormer=Yes, Mixing Type=Conv, Params (M)=95, Resolution=3842022.10 | 85.5 | 50.3 | — | |
| ConvNeXt-LMetaFormer=No, Mixing Type=Conv, Params (M)=198, Resolution=3842022.10 | 85.5 | 101 | — | |
| CAFormer-B36MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=99, Resolution=2242022.10 | 85.5 | 23.2 | — | |
| ConvFormer-S36MetaFormer=Yes, Mixing Type=Conv, Params (M)=40, Resolution=3842022.10 | 85.4 | 22.4 | — | |
| CSWin-BMetaFormer=Yes, Mixing Type=Attn, Params (M)=78, Resolution=3842022.10 | 85.4 | 47 | — | |
| MViTv2-LMetaFormer=Yes, Mixing Type=Attn, Params (M)=218, Resolution=2242022.10 | 85.3 | 42.1 | — | |
| MaxViT-SMetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=69, Resolution=3842022.10 | 85.2 | 36.1 | — | |
| CAFormer-M36MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=56, Resolution=2242022.10 | 85.2 | 13.2 | — | |
| CoAtNet-1MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=42, Resolution=3842022.10 | 85.1 | 27.4 | — | |
| ConvNeXt-BMetaFormer=No, Mixing Type=Conv, Params (M)=89, Resolution=3842022.10 | 85.1 | 45 | — | |
| CAFormer-S18MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=26, Resolution=3842022.10 | 85 | 13.4 | — | |
| MaxViT-BMetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=120, Resolution=2242022.10 | 85 | 23.4 | — | |
| RepLKNet-31BMetaFormer=No, Mixing Type=Conv, Params (M)=79, Resolution=3842022.10 | 84.8 | 45.1 | — | |
| iFormer-LMetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=87, Resolution=2242022.10 | 84.8 | 14 | — | |
| ConvFormer-B36MetaFormer=Yes, Mixing Type=Conv, Params (M)=100, Resolution=2242022.10 | 84.8 | 22.6 | — | |
| iFormer-SMetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=20, Resolution=3842022.10 | 84.6 | 16.1 | — | |
| CAFormer-S36MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=39, Resolution=2242022.10 | 84.5 | 8 | — | |
| ConvFormer-M36MetaFormer=Yes, Mixing Type=Conv, Params (M)=57, Resolution=2242022.10 | 84.5 | 12.8 | — | |
| Swin-BMetaFormer=Yes, Mixing Type=Attn, Params (M)=88, Resolution=3842022.10 | 84.5 | 47.1 | — | |
| MaxViT-SMetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=69, Resolution=2242022.10 | 84.5 | 11.7 | — | |
| CoAtNet-3MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=168, Resolution=2242022.10 | 84.5 | 34.7 | — | |
| ConvFormer-S18MetaFormer=Yes, Mixing Type=Conv, Params (M)=27, Resolution=3842022.10 | 84.4 | 11.6 | — | |
| MViTv2-BMetaFormer=Yes, Mixing Type=Attn, Params (M)=52, Resolution=2242022.10 | 84.4 | 10.2 | — | |
| ConvNeXt-LMetaFormer=No, Mixing Type=Conv, Params (M)=198, Resolution=2242022.10 | 84.3 | 34.4 | — | |
| VAN-B4MetaFormer=Yes, Mixing Type=Conv, Params (M)=60, Resolution=2242022.10 | 84.2 | 12.2 | — | |
| CSWin-BMetaFormer=Yes, Mixing Type=Attn, Params (M)=78, Resolution=2242022.10 | 84.2 | 15 | — | |
| ConvFormer-S36MetaFormer=Yes, Mixing Type=Conv, Params (M)=40, Resolution=2242022.10 | 84.1 | 7.6 | — | |
| CoAtNet-2MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=75, Resolution=2242022.10 | 84.1 | 15.7 | — | |
| SLaK-BMetaFormer=Yes, Mixing Type=Conv, Params (M)=95, Resolution=2242022.10 | 84 | 17.1 | — | |
| CoAtNet-0MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=25, Resolution=3842022.10 | 83.9 | 13.4 | — | |
| VAN-B3MetaFormer=Yes, Mixing Type=Conv, Params (M)=45, Resolution=2242022.10 | 83.9 | 9 | — | |
| UniFormer-BMetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=50, Resolution=2242022.10 | 83.9 | 8.3 | — | |
| ConvNeXt-BMetaFormer=No, Mixing Type=Conv, Params (M)=89, Resolution=2242022.10 | 83.8 | 15.4 | — | |
| CAFormer-S18MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=26, Resolution=2242022.10 | 83.6 | 4.1 | — | |
| CSWin-SMetaFormer=Yes, Mixing Type=Attn, Params (M)=35, Resolution=2242022.10 | 83.6 | 6.9 | — | |
| MViTv2-SMetaFormer=Yes, Mixing Type=Attn, Params (M)=35, Resolution=2242022.10 | 83.6 | 7 | — | |
| RepLKNet-31BMetaFormer=No, Mixing Type=Conv, Params (M)=79, Resolution=2242022.10 | 83.5 | 15.3 | — | |
| Swin-BMetaFormer=Yes, Mixing Type=Attn, Params (M)=88, Resolution=2242022.10 | 83.5 | 15.4 | — | |
| Dual-ViT-SMetaFormer=Yes, Mixing Type=Attn, Params (M)=25, Resolution=2242022.10 | 83.4 | 4.8 | — | |
| iFormer-SMetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=20, Resolution=2242022.10 | 83.4 | 4.8 | — | |
| T2T-ViT-14MetaFormer=Yes, Mixing Type=Attn, Params (M)=22, Resolution=3842022.10 | 83.3 | 17.1 | — | |
| CoAtNet-1MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=42, Resolution=2242022.10 | 83.3 | 8.4 | — | |
| ConvNeXt-SMetaFormer=No, Mixing Type=Conv, Params (M)=50, Resolution=2242022.10 | 83.1 | 8.7 | — | |
| DeiT-BMetaFormer=Yes, Mixing Type=Attn, Params (M)=86, Resolution=3842022.10 | 83.1 | 55.4 | — | |
| ConvFormer-S18MetaFormer=Yes, Mixing Type=Conv, Params (M)=27, Resolution=2242022.10 | 83 | 3.9 | — | |
| Swin-SMetaFormer=Yes, Mixing Type=Attn, Params (M)=50, Resolution=2242022.10 | 83 | 8.7 | — | |
| UniFormer-SMetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=22, Resolution=2242022.10 | 82.9 | 3.6 | — | |
| VAN-B2MetaFormer=Yes, Mixing Type=Conv, Params (M)=27, Resolution=2242022.10 | 82.8 | 5 | — | |
| CSWin-TMetaFormer=Yes, Mixing Type=Attn, Params (M)=23, Resolution=2242022.10 | 82.7 | 4.3 | — | |
| RegNetY-32GMetaFormer=No, Mixing Type=Conv, Params (M)=145, Resolution=2242022.10 | 82.4 | 32.3 | — | |
| MViTv2-TMetaFormer=Yes, Mixing Type=Attn, Params (M)=24, Resolution=2242022.10 | 82.3 | 4.7 | — | |
| T2T-ViT-24MetaFormer=Yes, Mixing Type=Attn, Params (M)=64, Resolution=2242022.10 | 82.3 | 13.8 | — | |
| RegNetY-16GMetaFormer=No, Mixing Type=Conv, Params (M)=84, Resolution=2242022.10 | 82.2 | 15.9 | — | |
| ConvNeXt-TMetaFormer=No, Mixing Type=Conv, Params (M)=29, Resolution=2242022.10 | 82.1 | 4.5 | — | |
| RegNetY-8GMetaFormer=No, Mixing Type=Conv, Params (M)=39, Resolution=2242022.10 | 82.1 | 8 | — | |
| T2T-ViT-19MetaFormer=Yes, Mixing Type=Attn, Params (M)=39, Resolution=2242022.10 | 81.9 | 8.5 | — | |
| RSB-ResNet-152MetaFormer=No, Mixing Type=Conv, Params (M)=60, Resolution=2242022.10 | 81.8 | 11.6 | — | |
| DeiT-BMetaFormer=Yes, Mixing Type=Attn, Params (M)=86, Resolution=2242022.10 | 81.8 | 17.5 | — | |
| CoAtNet-0MetaFormer=Yes, Mixing Type=Conv + Attn, Params (M)=25, Resolution=2242022.10 | 81.6 | 4.2 | — | |
| T2T-ViT-14MetaFormer=Yes, Mixing Type=Attn, Params (M)=22, Resolution=2242022.10 | 81.5 | 4.8 | — | |
| RegNetY-4GMetaFormer=No, Mixing Type=Conv, Params (M)=21, Resolution=2242022.10 | 81.3 | 4 | — | |
| Swin-TMetaFormer=Yes, Mixing Type=Attn, Params (M)=29, Resolution=2242022.10 | 81.3 | 4.5 | — | |
| RSB-ResNet-101MetaFormer=No, Mixing Type=Conv, Params (M)=45, Resolution=2242022.10 | 81.3 | 7.9 | — | |
| RSB-ResNet-50MetaFormer=No, Mixing Type=Conv, Params (M)=26, Resolution=2242022.10 | 79.8 | 4.1 | — | |
| DeiT-SMetaFormer=Yes, Mixing Type=Attn, Params (M)=22, Resolution=2242022.10 | 79.8 | 4.6 | — | |
| ScalaParam=22M, width ratio (r)=1, Backbone=DeiT [29]2024.12 | 76.1 | — | 4.6 | |
| Separate TrainingParam=43M, width ratio (r)=1, Backbone=DeiT [29]2024.12 | 75 | — | 4.6 | |
| ScalaParam=22M, width ratio (r)=0.75, Backbone=DeiT [29]2024.12 | 73.3 | — | 2.7 | |
| US-NetParam=22M, width ratio (r)=1, Backbone=DeiT [29]2024.12 | 73.2 | — | 4.6 | |
| Separate TrainingParam=43M, width ratio (r)=0.75, Backbone=DeiT [29]2024.12 | 70.7 | — | 2.7 | |
| AutoFormerParam=22M, width ratio (r)=1, Backbone=DeiT [29]2024.12 | 69.8 | — | 4.6 | |
| AutoFormerParam=22M, width ratio (r)=0.75, Backbone=DeiT [29]2024.12 | 69.5 | — | 2.7 | |
| ScalaParam=22M, width ratio (r)=0.5, Backbone=DeiT [29]2024.12 | 68.3 | — | 1.3 | |
| US-NetParam=22M, width ratio (r)=0.75, Backbone=DeiT [29]2024.12 | 66.9 | — | 2.7 | |
| AutoFormerParam=22M, width ratio (r)=0.5, Backbone=DeiT [29]2024.12 | 65.6 | — | 1.3 | |
| Separate TrainingParam=43M, width ratio (r)=0.5, Backbone=DeiT [29]2024.12 | 65.1 | — | 1.3 | |
| US-NetParam=22M, width ratio (r)=0.5, Backbone=DeiT [29]2024.12 | 60.1 | — | 1.3 | |
| ScalaParam=22M, width ratio (r)=0.25, Backbone=DeiT [29]2024.12 | 58.7 | — | 0.4 | |
| US-NetParam=22M, width ratio (r)=0.25, Backbone=DeiT [29]2024.12 | 52.7 | — | 0.4 | |
| AutoFormerParam=22M, width ratio (r)=0.25, Backbone=DeiT [29]2024.12 | 50.8 | — | 0.4 | |
| Separate TrainingParam=43M, width ratio (r)=0.25, Backbone=DeiT [29]2024.12 | 45.8 | — | 0.4 |