Image Classification on CIFAR100 (Accuracy)
92AccuracyMLCD
Evaluation Results
| Method | Links | |
|---|---|---|
| MLCDPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.07 | 92 | |
| EfficientNet-B5Params (M)=302022.02 | 91.1 | |
| DeiT-BParams (M)=86.62022.02 | 90.8 | |
| ViTAE-SParams (M)=23.62022.02 | 90.8 | |
| UNICOMPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.07 | 90.8 | |
| LIGOTraining mode=Training from the Pretrained Model: M(12,384) → M(12,768), FLOPs (x 1e18)=5.7, Ratio (Saving)=55.7%2023.10 | 90.52 | |
| bert2BERTTraining mode=Training from the Pretrained Model: M(12,384) → M(12,768), FLOPs (x 1e18)=4.6, Ratio (Saving)=64.4%2023.10 | 90.47 | |
| MangoTraining mode=Training from the Pretrained Model: M(12,384) → M(12,768), FLOPs (x 1e18)=3.0, Ratio (Saving)=76.4%2023.10 | 90.23 | |
| ScratchTraining mode=Training from Scratch, FLOPs (x 1e18)=12.92023.10 | 90.22 | |
| StackBERTTraining mode=Training from Scratch, FLOPs (x 1e18)=11.3, Ratio (Saving)=12.6%2023.10 | 90.1 | |
| GoGModel=DeiT-B2024.03 | 89.87 | |
| Greedy SoupModel=DeiT-B2024.03 | 89.68 | |
| GoUModel=DeiT-B2024.03 | 89.68 | |
| FGG(best)Model=DeiT-B2024.03 | 89.63 | |
| GS(best)Model=DeiT-B2024.03 | 89.19 | |
| T2T-ViT-14Params (M)=21.52022.02 | 88.4 | |
| OpenCLIPPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.07 | 87.9 | |
| BOFT (b=2, m=2)Backbone=ViT-L, #Params=0.20M2024.05 | 87.84 | |
| BOFT (b=4, m=1)Backbone=ViT-L, #Params=0.30M2024.05 | 87.72 | |
| CLIPPre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.07 | 87.5 | |
| SVFT^B (r=8)Backbone=ViT-L, #Params=2.50M2024.05 | 87.26 | |
| ViT-g/14N-shot=10, Probe=linear regression2023.05 | 87.2 | |
| CLIP+Pre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.07 | 87.2 | |
| ViT-B/16Params (M)=86.52022.02 | 87.1 | |
| SVFT^B (r=4)Backbone=ViT-L, #Params=1.32M2024.05 | 87.1 | |
| VeRABackbone=ViT-L, #Params=61.4K2024.05 | 86.77 | |
| SVFT^PBackbone=ViT-L, #Params=49.2K2024.05 | 86.74 | |
| SoViT-400m/14N-shot=10, Probe=linear regression2023.05 | 86.7 | |
| SVFT^B (r=2)Backbone=ViT-L, #Params=0.74M2024.05 | 86.59 | |
| Full-FTBackbone=ViT-L, #Params=303.3M2024.05 | 86.56 | |
| ViT-L/16Params (M)=304.32022.02 | 86.4 | |
| X-FM_baseLinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.01 | 86.2 | |
| ViTAE-TParams (M)=4.82022.02 | 86 | |
| LoRA (r=8)Backbone=ViT-L, #Params=0.35M2024.05 | 86 | |
| LoRA (r=1)Backbone=ViT-L, #Params=0.44M2024.05 | 85.97 | |
| DOT-CBM2025.05 | 85.83 | |
| SVFT^B (r=8)Backbone=ViT-B, #Params=0.94M2024.05 | 85.69 | |
| BOFT (b=2, m=2)Backbone=ViT-B, #Params=0.07M2024.05 | 85.55 | |
| BOFT (b=4, m=1)Backbone=ViT-B, #Params=0.11M2024.05 | 85.54 | |
| Uniform SoupModel=DeiT-B2024.03 | 85.51 | |
| Full-FTBackbone=ViT-B, #Params=85.8M2024.05 | 85.35 | |
| DoRA (r=8)Backbone=ViT-B, #Params=1.41M2024.05 | 85.03 | |
| LoRA (r=1)Backbone=ViT-B, #Params=0.16M2024.05 | 84.86 | |
| GoGModel=ResNet502024.03 | 84.77 | |
| SparseCBM2025.05 | 84.75 | |
| SVFT^B (r=2)Backbone=ViT-B, #Params=0.28M2024.05 | 84.72 | |
| CoopCBM2025.05 | 84.66 | |
| FGG(best)Model=ResNet502024.03 | 84.64 | |
| GoUModel=ResNet502024.03 | 84.57 | |
| DoRA (r=1)Backbone=ViT-B, #Params=0.25M2024.05 | 84.46 | |
| LoRA (r=8)Backbone=ViT-B, #Params=1.32M2024.05 | 84.41 | |
| FLIPPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 84.1 | |
| LaBo2025.05 | 84.1 | |
| DoRA (r=1)Backbone=ViT-L, #Params=0.66M2024.05 | 84.06 | |
| Greedy SoupModel=ResNet502024.03 | 84.01 | |
| SVFT^PBackbone=ViT-B, #Params=18.5K2024.05 | 83.85 | |
| GS(best)Model=ResNet502024.03 | 83.8 | |
| MLCDPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 83.7 | |
| DoRA (r=8)Backbone=ViT-L, #Params=3.76M2024.05 | 83.55 | |
| VeRABackbone=ViT-B, #Params=24.6K2024.05 | 83.38 | |
| OmniVLImg-text pairs=14M*, Evaluation protocol=Linear probing2022.09 | 83.2 | |
| SVFT^B (r=4)Backbone=ViT-B, #Params=0.50M2024.05 | 83.17 | |
| CLIPLinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.01 | 83.1 | |
| CLIP-ViT-B/16Img-text pairs=400M, Evaluation protocol=Linear probing2022.09 | 83.1 | |
| HeadBackbone=ViT-L2024.05 | 82.95 | |
| BASIC-LBackbone=BASIC-L2021.11 | 82.3 | |
| Stable SpikeArchitecture=ResNet-19, T=42026.03 | 82.29 | |
| ViT-L/16N-shot=10, Probe=linear regression2023.05 | 82.2 | |
| STAA-SNNArchitecture=ResNet-19, T=42026.03 | 82.05 | |
| Weak2StrongArchitecture=ResNet-19, T=42026.03 | 82.02 | |
| DeepTAGEArchitecture=ResNet-19, T=42026.03 | 81.39 | |
| CEM2025.05 | 81.23 | |
| QKFormerArchitecture=QKFormer, T=42026.03 | 81.15 | |
| ALBEFImg-text pairs=14M, Evaluation protocol=Linear probing2022.09 | 80.8 | |
| RateBPArchitecture=ResNet-19, T=42026.03 | 80.71 | |
| ResNet-50 + DASBackbone=ResNet-50, Attention=DAS2023.11 | 80.54 | |
| TS-SNNArchitecture=ResNet-19, T=22026.03 | 80.28 | |
| DaVinciLinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.01 | 80.1 | |
| SNN-ViTArchitecture=SNN-ViT, T=42026.03 | 80.1 | |
| Vanilla-CBM2025.05 | 80.04 | |
| Linear Probing2024.04 | 80.03 | |
| BLIPImg-text pairs=14M, Evaluation protocol=Linear probing2022.09 | 80 | |
| SWIN V2Resolution=64 x 64, Training Resolution=64 x 642022.05 | 79.95 | |
| SeTaBackbone=ResNet18, Pruned %=30%2025.03 | 79.4 | |
| CLIPEvaluation Protocol=10-shot linear evaluation, Model Scale=L/142023.06 | 79.4 | |
| SWEETModel Scale=L6H62026.01 | 79.36 | |
| Fully SupervisedData Partitioning=Centralized2021.06 | 79.3 | |
| ResNet-50 + Triplet AttentionBackbone=ResNet-50, Attention=Triplet Attention2023.11 | 79.22 | |
| InfoBatchBackbone=ResNet18, Pruned %=30%, Training Mode=Same number of epochs2025.03 | 79.2 | |
| VIT + TLBResolution=64 x 64, Training Resolution=64 x 642022.05 | 79.17 | |
| ResNet-18 + DASBackbone=ResNet-18, Attention=DAS2023.11 | 79.04 | |
| SeTaBackbone=ResNet18, Pruned %=50%2025.03 | 79 | |
| RényiCLEvaluation Protocol=linear evaluation, Multi-crop Augmentation=true2022.08 | 78.9 | |
| Stable SpikeArchitecture=ResNet-18, T=42026.03 | 78.83 | |
| RényiCLEvaluation Protocol=linear evaluation, Multi-crop Augmentation=false2022.08 | 78.8 | |
| ResNet-18 + Triplet AttentionBackbone=ResNet-18, Attention=Triplet Attention2023.11 | 78.55 | |
| ReverB-SNNArchitecture=ResNet-19, T=22026.03 | 78.46 | |
| RateBPArchitecture=ResNet-18, T=42026.03 | 78.26 | |
| ResNet-18Backbone=ResNet-182023.11 | 78.25 | |
| HeadBackbone=ViT-B2024.05 | 78.25 |