Image Classification on ImageNet (val) (Accuracy and Efficiency)
89.22Top-1 AccuracyGIT2
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GIT2Resolution=384, Setting=Fine-tuning2023.05 | 89.22 | — | — | — | — | — | |
| PaLI-XResolution=756, Setting=Fine-tuning, Steps=2.2x more steps2023.05 | 89.19 | — | — | — | — | — | |
| PaLI-XResolution=756, Setting=Fine-tuning2023.05 | 88.82 | — | — | — | — | — | |
| PaLI-XResolution=224, Setting=Fine-tuning2023.05 | 88.22 | — | — | — | — | — | |
| PaLIParameters=17B, Resolution=224, Setting=Fine-tuning2023.05 | 86.13 | — | — | — | — | — | |
| PaLIParameters=3B, Resolution=224, Setting=Fine-tuning2023.05 | 85.11 | — | — | — | — | — | |
| HALP-60%Backbone=ResNet1012022.10 | 78.3 | — | 4.3 | — | 847 | 1.37 | |
| HALP-50%Backbone=ResNet1012022.10 | 77.8 | — | 3.6 | — | 994 | 1.6 | |
| HALP-80%Backbone=ResNet50, Baseline=EagleEye [32]2022.10 | 77.5 | 93.6 | 3 | — | 1,203 | 1.18 | |
| No PruningBackbone=ResNet1012022.10 | 77.4 | — | 7.8 | — | 620 | 1 | |
| Taylor-75%Backbone=ResNet1012022.10 | 77.4 | — | 4.7 | — | 750 | 1.21 | |
| Flamingo-80BShots=5-shot2023.05 | 77.3 | — | — | — | — | — | |
| HALP-80%Backbone=ResNet502022.10 | 77.2 | 93.47 | 3.1 | — | 1,256 | 1.23 | |
| No PruningBackbone=ResNet50, Baseline=EagleEye [32]2022.10 | 77.2 | 93.7 | 4.1 | — | 1,019 | 1 | |
| HALP-40%Backbone=ResNet1012022.10 | 77.2 | — | 2.7 | — | 1,180 | 1.9 | |
| EagleEye-3GBackbone=ResNet50, Baseline=EagleEye [32]2022.10 | 77.1 | 93.37 | 3 | — | 1,165 | 1.14 | |
| HALP-55%Backbone=ResNet50, Baseline=EagleEye [32]2022.10 | 76.6 | 93.16 | 2.1 | — | 1,672 | 1.64 | |
| HALP-55%Backbone=ResNet502022.10 | 76.5 | 93.05 | 2 | — | 1,630 | 1.6 | |
| HALP-30%Backbone=ResNet1012022.10 | 76.5 | — | 2 | — | 1,521 | 2.45 | |
| CHIPBackbone=ResNet50, Target FLOPs=2.2G2022.10 | 76.4 | 93.05 | 2.2 | — | 1,345 | 1.32 | |
| EagleEye-2GBackbone=ResNet50, Baseline=EagleEye [32]2022.10 | 76.4 | 92.89 | 2.1 | — | 1,471 | 1.44 | |
| GReg-1Backbone=ResNet502022.10 | 76.3 | — | 2.7 | — | 1,171 | 1.15 | |
| No PruningBackbone=ResNet502022.10 | 76.2 | 92.87 | 4.1 | — | 1,019 | 1 | |
| MetaPruningBackbone=ResNet502022.10 | 76.2 | — | 3 | — | — | — | |
| GBNBackbone=ResNet502022.10 | 76.2 | 92.83 | 2.4 | — | — | — | |
| CHIPBackbone=ResNet50, Target FLOPs=2.1G2022.10 | 76.2 | 92.91 | 2.1 | — | — | — | |
| AutoSlimBackbone=ResNet502022.10 | 76 | — | 3 | — | 1,215 | 1.14 | |
| Taylor-55%Backbone=ResNet1012022.10 | 76 | — | 2.9 | — | 908 | 1.47 | |
| ThiNet-70Backbone=ResNet502022.10 | 75.8 | 90.67 | 2.9 | — | — | — | |
| AutoSlimBackbone=ResNet50, Target FLOPs=2.0G2022.10 | 75.6 | — | 2 | — | 1,592 | 1.56 | |
| CAIEBackbone=ResNet50, Target FLOPs=2.2G2022.10 | 75.6 | — | 2.2 | — | — | — | |
| LEGRBackbone=ResNet502022.10 | 75.6 | 92.7 | 2.4 | — | — | — | |
| MetaPruningBackbone=ResNet50, Target FLOPs=2.0G2022.10 | 75.4 | — | 2 | — | 1,604 | 1.58 | |
| GReg-2Backbone=ResNet50, Target FLOPs=1.8G2022.10 | 75.4 | — | 1.8 | — | 1,414 | 1.39 | |
| 0.75x ResNet50Backbone=ResNet502022.10 | 74.8 | — | 2.3 | — | 1,467 | 1.44 | |
| ThiNet-50Backbone=ResNet502022.10 | 74.7 | 90.02 | 2.1 | — | — | — | |
| HALP-30%Backbone=ResNet50, Baseline=EagleEye [32]2022.10 | 74.5 | 91.87 | 1 | — | 2,597 | 2.55 | |
| HALP-30%Backbone=ResNet502022.10 | 74.3 | 91.81 | 1 | — | 2,755 | 2.7 | |
| EagleEye-1GBackbone=ResNet50, Baseline=EagleEye [32]2022.10 | 74.2 | 91.77 | 1 | — | 2,429 | 2.38 | |
| AutoSlimBackbone=ResNet50, Target FLOPs=1.0G2022.10 | 74 | — | 1 | — | 2,390 | 2.45 | |
| CAIEBackbone=ResNet50, Target FLOPs=1.3G2022.10 | 73.9 | — | 1.3 | — | — | — | |
| GReg-2Backbone=ResNet50, Target FLOPs=1.3G2022.10 | 73.9 | — | 1.3 | — | 1,514 | 1.49 | |
| MetaPruningBackbone=ResNet50, Target FLOPs=1.0G2022.10 | 73.4 | — | 1 | — | 2,381 | 2.34 | |
| CHIPBackbone=ResNet50, Target FLOPs=1.0G2022.10 | 73.3 | 91.48 | 1 | — | 2,369 | 2.32 | |
| No PruningBackbone=MobileNet-V12022.10 | 72.6 | — | — | 569 | 3,415 | 1 | |
| HALP-30%Backbone=VGG-162022.10 | 72.3 | — | 4.6 | — | 1,498 | 2.42 | |
| HALP-75%Backbone=MobileNet-V22022.10 | 72.2 | — | — | 249 | 4,110 | 1.33 | |
| PaLIParameters=17B, Shots=0-shot, Resolution=2242023.05 | 72.11 | — | — | — | — | — | |
| ThiNet-30Backbone=ResNet502022.10 | 72.1 | 88.3 | 1.2 | — | — | — | |
| No PruningBackbone=MobileNet-V22022.10 | 72.1 | — | — | 301 | 3,080 | 1 | |
| 0.50x ResNet50Backbone=ResNet502022.10 | 72 | — | 1.1 | — | 2,498 | 2.45 | |
| Flamingo-80BShots=1-shot2023.05 | 71.9 | — | — | — | — | — | |
| No PruningBackbone=VGG-162022.10 | 71.6 | — | 15.5 | — | 766 | 1 | |
| GDPBackbone=MobileNet-V12022.10 | 71.3 | — | — | 287 | — | — | |
| HALP-60%Backbone=MobileNet-V12022.10 | 71.3 | — | — | 297 | 5,754 | 1.68 | |
| FBS-3xBackbone=VGG-162022.10 | 71.2 | — | 5.1 | — | — | — | |
| PaLI-XShots=0-shot, Resolution=2242023.05 | 71.16 | — | — | — | — | — | |
| MetaPruningBackbone=MobileNet-V1, Target FLOPS=316M2022.10 | 70.9 | — | — | 316 | 4,838 | 1.42 | |
| EagleEyeBackbone=MobileNet-V12022.10 | 70.9 | — | — | 284 | 5,020 | 1.47 | |
| HALP-20%Backbone=VGG-162022.10 | 70.8 | — | 2.8 | — | 1,958 | 5.49 | |
| FBS-5xBackbone=VGG-162022.10 | 70.5 | — | 3 | — | — | — | |
| AMCBackbone=MobileNet-V12022.10 | 70.5 | — | — | 285 | 4,857 | 1.42 | |
| HALP-60%Backbone=MobileNet-V22022.10 | 70.4 | — | — | 183 | 5,668 | 1.84 | |
| NetAdaptBackbone=MobileNet-V12022.10 | 69.1 | — | — | 284 | — | — | |
| 0.75x MobileNetV1Backbone=MobileNet-V12022.10 | 68.4 | — | — | 325 | 4,678 | 1.37 | |
| HALP-42%Backbone=MobileNet-V12022.10 | 68.3 | — | — | 171 | 7,940 | 2.32 | |
| AutoSlimBackbone=MobileNet-V12022.10 | 67.9 | — | — | 150 | 7,743 | 2.27 | |
| MetaPruningBackbone=MobileNet-V1, Target FLOPS=142M2022.10 | 66.1 | — | — | 142 | 7,050 | 2.06 |