Image Classification on CIFAR10 (Top-1 Accuracy, Params, FLOPs)
98Top-1 AccCLIP
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CLIPevaluation=linear probing2022.08 | 98 | — | — | — | — | |
| LOUPEevaluation=linear probing2022.08 | 97.6 | — | — | — | — | |
| FILIPBackbone=Swin-L, Zero-shot=true, Pre-training Dataset=Wukong2022.02 | 95.5 | — | — | — | — | |
| WukongBackbone=ViT-L, Zero-shot=true, Pre-training Dataset=Wukong2022.02 | 95.4 | — | — | — | — | |
| WukongBackbone=Swin-L, Zero-shot=true, Pre-training Dataset=Wukong2022.02 | 95.3 | — | — | — | — | |
| DECORE-175Backbone=DenseNet-40, Penalty (lambda)=1752021.06 | 94.85 | — | 20.7 | — | 19.1 | |
| DenseNet-40Backbone=DenseNet-402021.06 | 94.81 | — | 0 | — | 0 | |
| Liu et al.Backbone=DenseNet-40, Pruned rate=40%2021.06 | 94.81 | — | 36.5 | — | 32.8 | |
| CLIPBackbone=Swin-L, Zero-shot=true, Pre-training Dataset=Wukong2022.02 | 94.8 | — | — | — | — | |
| Full-precisionBackbone=ResNet-18, Bit-width (W/A)=32/32, Storage (Mbit)=351.5, BOPs (x10^8)=350.32023.03 | 94.8 | — | — | — | — | |
| DECORE-115Backbone=DenseNet-40, Penalty (lambda)=1152021.06 | 94.59 | — | 46 | — | 39.4 | |
| RényiCLProtocol=linear evaluation2022.08 | 94.4 | — | — | — | — | |
| GAL-0.01Backbone=DenseNet-40, Sparsity factor (GAL-X)=0.012021.06 | 94.29 | — | 35.6 | — | 35.3 | |
| HRank-1Backbone=DenseNet-40, Compression rate setting (HRANK-N)=12021.06 | 94.24 | — | 36.5 | — | 40.8 | |
| CLIPBackbone=ViT-L, Zero-shot=true, Pre-training Dataset=Wukong2022.02 | 94.1 | — | — | — | — | |
| Full-precisionBackbone=VGG-small, Bit-width (W/A)=32/32, Storage (Mbit)=146.2, BOPs (x10^8)=386.62023.03 | 94.1 | — | — | — | — | |
| DECORE-70Backbone=DenseNet-40, Penalty (lambda)=702021.06 | 94.04 | — | 65 | — | 54.7 | |
| NNCLRProtocol=linear evaluation2022.08 | 93.7 | — | — | — | — | |
| HRank-2Backbone=DenseNet-40, Compression rate setting (HRANK-N)=22021.06 | 93.68 | — | 53.8 | — | 61 | |
| SupervisedProtocol=linear evaluation2022.08 | 93.6 | — | — | — | — | |
| GAL-0.05Backbone=DenseNet-40, Sparsity factor (GAL-X)=0.052021.06 | 93.53 | — | 56.7 | — | 54.7 | |
| Zhao et al.Backbone=DenseNet-402021.06 | 93.16 | — | 59.7 | — | 44.8 | |
| ReActNetBackbone=ResNet-18, Bit-width (W/A)=1/1, Storage (Mbit)=11.0, BOPs (x10^8)=5.472023.03 | 92.3 | — | — | — | — | |
| RBNNBackbone=ResNet-18, Bit-width (W/A)=1/1, Storage (Mbit)=11.0, BOPs (x10^8)=5.472023.03 | 92.2 | — | — | — | — | |
| SparksBackbone=ResNet-18, Bit-width (W/A)=0.78/1, Storage (Mbit)=8.57, BOPs (x10^8)=3.962023.03 | 92.2 | — | — | — | — | |
| SparksBackbone=ResNet-18, Bit-width (W/A)=0.67/1, Storage (Mbit)=7.32, BOPs (x10^8)=2.972023.03 | 92 | — | — | — | — | |
| SLBBackbone=VGG-small, Bit-width (W/A)=1/1, Storage (Mbit)=4.57, BOPs (x10^8)=6.032023.03 | 92 | — | — | — | — | |
| SparksBackbone=VGG-small, Bit-width (W/A)=0.78/1, Storage (Mbit)=3.55, BOPs (x10^8)=3.462023.03 | 91.7 | — | — | — | — | |
| SparksBackbone=VGG-small, Bit-width (W/A)=0.67/1, Storage (Mbit)=3.05, BOPs (x10^8)=1.942023.03 | 91.6 | — | — | — | — | |
| IR-NetBackbone=ResNet-18, Bit-width (W/A)=1/1, Storage (Mbit)=11.0, BOPs (x10^8)=5.472023.03 | 91.5 | — | — | — | — | |
| SparksBackbone=ResNet-18, Bit-width (W/A)=0.56/1, Storage (Mbit)=6.10, BOPs (x10^8)=1.632023.03 | 91.5 | — | — | — | — | |
| BYOLProtocol=linear evaluation2022.08 | 91.3 | — | — | — | — | |
| RBNNBackbone=VGG-small, Bit-width (W/A)=1/1, Storage (Mbit)=4.57, BOPs (x10^8)=6.032023.03 | 91.3 | — | — | — | — | |
| SparksBackbone=VGG-small, Bit-width (W/A)=0.56/1, Storage (Mbit)=2.54, BOPs (x10^8)=1.132023.03 | 91.3 | — | — | — | — | |
| FleXORBackbone=ResNet-18, Bit-width (W/A)=0.80/1, Storage (Mbit)=8.80, BOPs (x10^8)=5.472023.03 | 90.9 | — | — | — | — | |
| SparksBackbone=ResNet-18, Bit-width (W/A)=0.44/1, Storage (Mbit)=4.88, BOPs (x10^8)=0.972023.03 | 90.8 | — | — | — | — | |
| SparksBackbone=VGG-small, Bit-width (W/A)=0.44/1, Storage (Mbit)=2.03, BOPs (x10^8)=0.742023.03 | 90.8 | — | — | — | — | |
| FILIPBackbone=ViT-L, Zero-shot=true, Pre-training Dataset=Wukong2022.02 | 90.6 | — | — | — | — | |
| SimCLRProtocol=linear evaluation2022.08 | 90.6 | — | — | — | — | |
| FleXORBackbone=VGG-small, Bit-width (W/A)=0.80/1, Storage (Mbit)=3.66, BOPs (x10^8)=6.032023.03 | 90.6 | — | — | — | — | |
| RADBackbone=ResNet-18, Bit-width (W/A)=1/1, Storage (Mbit)=11.0, BOPs (x10^8)=5.472023.03 | 90.5 | — | — | — | — | |
| IR-NetBackbone=VGG-small, Bit-width (W/A)=1/1, Storage (Mbit)=4.57, BOPs (x10^8)=6.032023.03 | 90.4 | — | — | — | — | |
| XNOR-NetBackbone=ResNet-18, Bit-width (W/A)=1/1, Storage (Mbit)=11.0, BOPs (x10^8)=5.472023.03 | 90.2 | — | — | — | — | |
| Bi-RealNetBackbone=ResNet-18, Bit-width (W/A)=1/1, Storage (Mbit)=11.0, BOPs (x10^8)=5.472023.03 | 90.2 | — | — | — | — | |
| RADBackbone=VGG-small, Bit-width (W/A)=1/1, Storage (Mbit)=4.57, BOPs (x10^8)=6.032023.03 | 90 | — | — | — | — | |
| FleXORBackbone=ResNet-18, Bit-width (W/A)=0.60/1, Storage (Mbit)=6.60, BOPs (x10^8)=5.472023.03 | 89.8 | — | — | — | — | |
| XNOR-NetBackbone=VGG-small, Bit-width (W/A)=1/1, Storage (Mbit)=4.57, BOPs (x10^8)=6.032023.03 | 89.8 | — | — | — | — | |
| CLIPBackbone=ViT-B, Zero-shot=true, Pre-training Dataset=Wukong2022.02 | 89.4 | — | — | — | — | |
| SLBFBackbone=VGG-small, Bit-width (W/A)=0.53/1, Storage (Mbit)=2.42, BOPs (x10^8)=3.172023.03 | 89.4 | — | — | — | — | |
| SLBFBackbone=ResNet-18, Bit-width (W/A)=0.55/1, Storage (Mbit)=6.05, BOPs (x10^8)=2.942023.03 | 89.3 | — | — | — | — | |
| FleXORBackbone=VGG-small, Bit-width (W/A)=0.60/1, Storage (Mbit)=2.74, BOPs (x10^8)=6.032023.03 | 89.2 | — | — | — | — | |
| LABBackbone=VGG-small, Bit-width (W/A)=1/1, Storage (Mbit)=4.57, BOPs (x10^8)=6.032023.03 | 87.7 | — | — | — | — | |
| WukongBackbone=ViT-B, Zero-shot=true, Pre-training Dataset=Wukong2022.02 | 87.1 | — | — | — | — | |
| FILIPBackbone=ViT-B, Zero-shot=true, Pre-training Dataset=Wukong2022.02 | 87 | — | — | — | — | |
| BriVLZero-shot=true, Pre-training Dataset=Own dataset2022.02 | 72.3 | — | — | — | — |