Image Classification on ImageNet-1k (Standard Accuracy)
84.52Top-1 AccuracyNAMEx-Full-Mom
Evaluation results
| Method | Links | ||
|---|---|---|---|
| NAMEx-Full-MomParams=50M, Momentum=true, Fine-tuning=true2025.10 | 84.52 | 98.11 | |
| NAMEx-MomParams=50M, Momentum=true, Fine-tuning=true2025.10 | 84.28 | 97.94 | |
| NAMEx-FullParams=50M, Momentum=false, Fine-tuning=true2025.10 | 84.27 | 97.94 | |
| NAMExParams=50M, Momentum=false, Fine-tuning=true2025.10 | 84.06 | 97.19 | |
| EP-CAMEx-MomParams=50M, Momentum=true, Fine-tuning=true2025.10 | 83.56 | 97.03 | |
| SwinTransformer-B (Dense)Backbone=Swin-B, MACs (G)=15.4, Throughput (images/s)=690.25, Params (M)=87.8, Speedup (x)=1.002026.07 | 83.5 | — | |
| CAMExParams=50M, Fine-tuning=true2025.10 | 83.29 | 96.95 | |
| EP-CAMExParams=50M, Momentum=false, Fine-tuning=true2025.10 | 83.23 | 96.93 | |
| SwinTransformer-S (Dense)Backbone=Swin-S, MACs (G)=8.7, Throughput (images/s)=1106.93, Params (M)=49.6, Speedup (x)=1.002026.07 | 83.2 | — | |
| SMoEParams=50M, Fine-tuning=true2025.10 | 83.15 | 96.71 | |
| SMEARParams=50M, Fine-tuning=true2025.10 | 83.15 | 96.91 | |
| HetDPTBackbone=Swin-B, MACs (G)=9.8, Throughput (images/s)=1106.95, Params (M)=55.3, Speedup (x)=1.602026.07 | 82.9 | — | |
| SAViTBackbone=Swin-B, MACs (G)=7.7, Throughput (images/s)=1056.08, Params (M)=33.0, Speedup (x)=1.532026.07 | 82.6 | — | |
| HetDPT (10 layers)Backbone=Swin-S, MACs (G)=7.3, Throughput (images/s)=1365.89, Params (M)=43.9, Speedup (x)=1.232026.07 | 82.4 | — | |
| WD-Pruning (0.2-24)Backbone=Swin-S, MACs (G)=6.8, Throughput (images/s)=1140.14, Params (M)=39.2, Speedup (x)=1.102026.07 | 82.2 | — | |
| HetDPT (8 layers)Backbone=DeiT-B, MACs (G)=11.8, Throughput (images/s)=1048.5, Params (M)=61.4, Speedup (x)=1.412026.07 | 82.1 | — | |
| WD-Pruning (0.2-12)Backbone=DeiT-B, MACs (G)=13.6, Throughput (images/s)=907.6, Params (M)=68.2, Speedup (x)=1.222026.07 | 81.9 | — | |
| DeiT-B (Dense)Backbone=DeiT-B, MACs (G)=17.6, Throughput (images/s)=741.5, Params (M)=86.6, Speedup (x)=1.002026.07 | 81.8 | — | |
| NOSE (5 layers)Backbone=DeiT-B, MACs (G)=14.6, Throughput (images/s)=894.7, Params (M)=75.5, Speedup (x)=1.212026.07 | 81.8 | — | |
| HetDPT (10 layers)Backbone=DeiT-B, MACs (G)=10.5, Throughput (images/s)=1169.7, Params (M)=54.9, Speedup (x)=1.582026.07 | 81.8 | — | |
| WD-Pruning (0.2-22)Backbone=Swin-S, MACs (G)=6.3, Throughput (images/s)=1272.97, Params (M)=34.4, Speedup (x)=1.152026.07 | 81.8 | — | |
| GTP-ViTBackbone=DeiT-B, MACs (G)=13.2, Throughput (images/s)=955.3, Params (M)=86.6, Speedup (x)=1.292026.07 | 81.5 | — | |
| NOSE (6 layers)Backbone=DeiT-B, MACs (G)=14.1, Throughput (images/s)=921.4, Params (M)=73.2, Speedup (x)=1.242026.07 | 81.5 | — | |
| TPSBackbone=DeiT-B, MACs (G)=12.9, Throughput (images/s)=993.9, Params (M)=89.5, Speedup (x)=1.342026.07 | 81.4 | — | |
| SGDBackbone=ResNet-101, Epoch=2002025.06 | 81.3 | — | |
| DynamicViTBackbone=DeiT-B, MACs (G)=11.2, Throughput (images/s)=1151.5, Params (M)=89.5, Speedup (x)=1.552026.07 | 81.3 | — | |
| Evo-ViTBackbone=DeiT-B, MACs (G)=11.3, Throughput (images/s)=1147.0, Params (M)=87.3, Speedup (x)=1.552026.07 | 81.3 | — | |
| EViTBackbone=DeiT-B, MACs (G)=11.2, Throughput (images/s)=1138.1, Params (M)=86.6, Speedup (x)=1.532026.07 | 81.3 | — | |
| MACBackbone=ResNet-101, Epoch=2002025.06 | 81.2 | — | |
| SMACBackbone=ResNet-101, Epoch=2002025.06 | 81.1 | — | |
| KFACBackbone=ResNet-101, Epoch=2002025.06 | 81.1 | — | |
| EVABackbone=ResNet-101, Epoch=2002025.06 | 81.1 | — | |
| FOOFBackbone=ResNet-101, Epoch=2002025.06 | 81 | — | |
| WD-Pruning (0.2-10)Backbone=DeiT-B, MACs (G)=10.8, Throughput (images/s)=1037.9, Params (M)=59.4, Speedup (x)=1.402026.07 | 80.9 | — | |
| ADAMWBackbone=ResNet-101, Epoch=2002025.06 | 80.6 | — | |
| ToMeBackbone=DeiT-B, MACs (G)=13.2, Throughput (images/s)=949.2, Params (M)=86.6, Speedup (x)=1.282026.07 | 80.6 | — | |
| HetDPT (6 layers)Backbone=DeiT-S, MACs (G)=3.3, Throughput (images/s)=2987.3, Params (M)=17.6, Speedup (x)=1.272026.07 | 80.1 | — | |
| FOOFBackbone=ResNet-101, Epoch=1002025.06 | 80 | — | |
| SMACBackbone=ResNet-50, Epoch=2002025.06 | 79.9 | — | |
| SMACBackbone=ResNet-101, Epoch=1002025.06 | 79.9 | — | |
| MACBackbone=ResNet-101, Epoch=1002025.06 | 79.8 | — | |
| DeiT-S (Dense)Backbone=DeiT-S, MACs (G)=4.6, Throughput (images/s)=2349.9, Params (M)=22.1, Speedup (x)=1.002026.07 | 79.8 | — | |
| NOSE (4 layers)Backbone=DeiT-S, MACs (G)=3.8, Throughput (images/s)=2774.8, Params (M)=20.1, Speedup (x)=1.182026.07 | 79.8 | — | |
| MACBackbone=ResNet-50, Epoch=2002025.06 | 79.7 | — | |
| SGDBackbone=ResNet-101, Epoch=1002025.06 | 79.7 | — | |
| FOOFBackbone=ResNet-50, Epoch=2002025.06 | 79.7 | — | |
| TPSBackbone=DeiT-S, MACs (G)=3.3, Throughput (images/s)=3051.2, Params (M)=22.8, Speedup (x)=1.302026.07 | 79.7 | — | |
| HetDPT (8 layers)Backbone=DeiT-S, MACs (G)=3.0, Throughput (images/s)=3275.0, Params (M)=15.9, Speedup (x)=1.392026.07 | 79.7 | — | |
| SGDBackbone=ResNet-50, Epoch=2002025.06 | 79.6 | — | |
| KFACBackbone=ResNet-101, Epoch=1002025.06 | 79.6 | — | |
| EVABackbone=ResNet-101, Epoch=1002025.06 | 79.6 | — | |
| NOSE (5 layers)Backbone=DeiT-S, MACs (G)=3.7, Throughput (images/s)=2875.3, Params (M)=19.5, Speedup (x)=1.222026.07 | 79.6 | — | |
| ToMeBackbone=DeiT-S, MACs (G)=3.3, Throughput (images/s)=2874.5, Params (M)=22.1, Speedup (x)=1.222026.07 | 79.5 | — | |
| GTP-ViTBackbone=DeiT-S, MACs (G)=3.5, Throughput (images/s)=2793.4, Params (M)=22.1, Speedup (x)=1.192026.07 | 79.5 | — | |
| EVABackbone=ResNet-50, Epoch=2002025.06 | 79.4 | — | |
| KFACBackbone=ResNet-50, Epoch=2002025.06 | 79.3 | — | |
| ADAMWBackbone=ResNet-50, Epoch=2002025.06 | 79.2 | — | |
| WD-Pruning (0.3-12)Backbone=DeiT-S, MACs (G)=3.1, Throughput (images/s)=2989.8, Params (M)=15.0, Speedup (x)=1.272026.07 | 78.6 | — | |
| FOOFBackbone=ResNet-50, Epoch=1002025.06 | 78.4 | — | |
| KFACBackbone=ResNet-50, Epoch=1002025.06 | 78.2 | — | |
| SMACBackbone=ResNet-50, Epoch=1002025.06 | 78.1 | — | |
| SGDBackbone=ResNet-50, Epoch=1002025.06 | 78.1 | — | |
| MACBackbone=ResNet-50, Epoch=1002025.06 | 78 | — | |
| ADAMWBackbone=ResNet-101, Epoch=1002025.06 | 77.9 | — | |
| EVABackbone=ResNet-50, Epoch=1002025.06 | 77.7 | — | |
| ADAMWBackbone=ResNet-50, Epoch=1002025.06 | 76.8 | — |