Image Classification on CIFAR10 (Acc %)
99.15Accuracy (%)Full Fine-tuning
Evaluation results
| Method | Links | |
|---|---|---|
| Full Fine-tuningBackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | 99.15 | |
| FFBackbone=ViT-Large, Method=FF, # Params=303.3M2026.05 | 99.15 | |
| GPartBackbone=ViT-Large, Method=GPart (Ours), # Params=144K2026.05 | 99.11 | |
| FourierFTBackbone=ViT-Large, Method=FourierFT, # Params=144K2026.05 | 99.1 | |
| FourierFTBackbone=ViT-Large, Method=FourierFT, # Params=480K2026.05 | 99.08 | |
| BaLoRABackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | 99.04 | |
| DoRABackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | 99.03 | |
| LoRABackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | 98.93 | |
| FFBackbone=ViT-Base, Method=FF, # Params=85.8M2026.05 | 98.92 | |
| Uni-LoRABackbone=ViT-Base, Method=Uni-LoRA, # Params=72K2026.05 | 98.77 | |
| GPartBackbone=ViT-Base, Method=GPart (Ours), # Params=72K2026.05 | 98.77 | |
| Uni-LoRABackbone=ViT-Large, Method=Uni-LoRA, # Params=144K2026.05 | 98.77 | |
| FourierFTBackbone=ViT-Base, Method=FourierFT, # Params=239K2026.05 | 98.69 | |
| FourierFTBackbone=ViT-Base, Method=FourierFT, # Params=72K2026.05 | 98.58 | |
| ViT-BAdaptation Strategy=Fine-tuning, Params=85.8M, Pre-training Dataset=ImageNet-1K2026.05 | 98.3 | |
| bViT-BAdaptation Strategy=Fine-tuning, Params=7.8M, Pre-training Dataset=ImageNet-1K2026.05 | 98.2 | |
| ViT-BAdaptation Strategy=LoRA, FFN, Params=738K, Pre-training Dataset=ImageNet-1K2026.05 | 97.9 | |
| PARA2026.04 | 97.89 | |
| ViT-BAdaptation Strategy=LoRA, QV, Params=296K, Pre-training Dataset=ImageNet-1K2026.05 | 97.8 | |
| LPBackbone=ViT-Large, Method=LP, # Params=02026.05 | 97.78 | |
| bViT-BAdaptation Strategy=LoRA, FFN, Params=62K, Pre-training Dataset=ImageNet-1K2026.05 | 97.7 | |
| LoRA2026.04 | 97.53 | |
| SAFformerParam=6.312026.05 | 97.5 | |
| bViT-BAdaptation Strategy=LoRA, QV, Params=25K, Pre-training Dataset=ImageNet-1K2026.05 | 97.5 | |
| STAA-SNN2026.05 | 97.14 | |
| Max-FormerParam=6.572026.05 | 97.04 | |
| bViT-BAdaptation Strategy=Time embedding tuning, Params=9K, Pre-training Dataset=ImageNet-1K2026.05 | 97 | |
| GoRA2026.04 | 96.91 | |
| Trasnformer (ANN)Param (M)=9.32, Time Step (T)=12024.03 | 96.73 | |
| MSViTParam=7.592026.05 | 96.53 | |
| FSTA-SNNParam=11.302026.05 | 96.52 | |
| DoRA2026.04 | 96.51 | |
| Spikinghash-MParam (M)=6.24, Time Step=42025.01 | 96.5 | |
| Burst+LIPoolingType=ANN2SNN, Architecture=ResNet-20, Timestep=642026.04 | 96.49 | |
| QB-LIFType=SNN Training, Architecture=ResNet-20, Timestep=42026.04 | 96.43 | |
| LPBackbone=ViT-Base, Method=LP, # Params=02026.05 | 96.41 | |
| SpikingResformer-TiParam (M)=10.79, Time Step=42025.01 | 96.24 | |
| SoRA2026.04 | 96.24 | |
| QB-LIFType=SNN Training, Architecture=ResNet-20, Timestep=22026.04 | 96.21 | |
| QKFormerParam (M)=6.74, Time Step=42025.01 | 96.18 | |
| QKFormerParam=6.742026.05 | 96.18 | |
| QKFormerParam (M)=6.74, Time Step (T)=42024.03 | 96.18 | |
| AdaLoRA2026.04 | 96.09 | |
| CMLParam (M)=9.32, Time Step (T)=42024.03 | 96.04 | |
| Spikinghash-SParam (M)=3.87, Time Step=42025.01 | 96.01 | |
| Spikingformer-CMLParam (M)=9.32, Time Step=42025.01 | 95.95 | |
| SpikingformerParam (M)=9.32, Time Step (T)=42024.03 | 95.81 | |
| SpikingformerParam (M)=9.32, Time Step=42025.01 | 95.61 | |
| Spike-driven TransformerParam (M)=9.32, Time Step=42025.01 | 95.6 | |
| S-TransformerParam=10.282026.05 | 95.6 | |
| S-TransformerParam (M)=10.28, Time Step (T)=42024.03 | 95.6 | |
| RecDis-SNNType=SNN Training, Architecture=ResNet-19, Timestep=42026.04 | 95.53 | |
| SpikformerParam=9.322026.05 | 95.51 | |
| SpikformerParam (M)=9.32, Time Step (T)=42024.03 | 95.51 | |
| QB-LIFType=SNN Training, Architecture=ResNet-20, Timestep=12026.04 | 95.4 | |
| VeCAFSampling Ratio=2%, Backbone=ViT, multi-run=true2026.06 | 95.27 | |
| TeacherBackbone=ResNet182026.04 | 95.21 | |
| SpikformerParam (M)=9.32, Time Step=42025.01 | 95.19 | |
| STAA-SNNType=SNN Training, Architecture=ResNet-20, Timestep=42026.04 | 95.03 | |
| ResNet-19 (ANN)Param (M)=12.63, Time Step (T)=12024.03 | 94.97 | |
| Ternary SpikeType=SNN Training, Architecture=ResNet-20, Timestep=42026.04 | 94.96 | |
| GLIFType=SNN Training, Architecture=ResNet-19, Timestep=42026.04 | 94.85 | |
| FSTA-SNNType=SNN Training, Architecture=ResNet-20, Timestep=42026.04 | 94.72 | |
| Ternary SpikeType=SNN Training, Architecture=ResNet-20, Timestep=22026.04 | 94.48 | |
| GLIFType=SNN Training, Architecture=ResNet-19, Timestep=22026.04 | 94.44 | |
| TETType=SNN Training, Architecture=ResNet-19, Timestep=42026.04 | 94.44 | |
| BEiT-B/16 + CLIPImg. encoder=BEiT-B/16, Txt. encoder=CLIP2026.05 | 94.4 | |
| STAA-SNNType=SNN Training, Architecture=ResNet-20, Timestep=22026.04 | 94.35 | |
| DspikeParam (M)=11.17, Time Step=62025.01 | 94.25 | |
| MLFType=SNN Training, Architecture=ResNet-19, Timestep=42026.04 | 94.25 | |
| L3FACTSampling Ratio=2%, Backbone=ViT2026.06 | 94.25 | |
| FSTA-SNNType=SNN Training, Architecture=ResNet-20, Timestep=22026.04 | 94.18 | |
| TETType=SNN Training, Architecture=ResNet-19, Timestep=22026.04 | 94.16 | |
| DA-LIFType=SNN Training, Architecture=ResNet-20, Timestep=42026.04 | 94.16 | |
| Grad-MimicNoise Level=0.4, Evaluation Protocol=linear probing, Backbone=ViT-B/162025.01 | 94.15 | |
| Rho-LossNoise Level=0.4, Evaluation Protocol=linear probing, Backbone=ViT-B/162025.01 | 94.07 | |
| FACTSampling Ratio=2%, Backbone=ViT2026.06 | 94.05 | |
| Grad-MimicNoise Level=0.5, Evaluation Protocol=linear probing, Backbone=ViT-B/162025.01 | 93.92 | |
| Rho-LossNoise Level=0.5, Evaluation Protocol=linear probing, Backbone=ViT-B/162025.01 | 93.83 | |
| Rho-LossNoise Level=0.6, Evaluation Protocol=linear probing, Backbone=ViT-B/162025.01 | 93.82 | |
| Grad-MimicNoise Level=0.6, Evaluation Protocol=linear probing, Backbone=ViT-B/162025.01 | 93.8 | |
| DA-LIFType=SNN Training, Architecture=ResNet-20, Timestep=22026.04 | 93.65 | |
| RecDis-SNNType=SNN Training, Architecture=ResNet-19, Timestep=22026.04 | 93.64 | |
| VeCAFSampling Ratio=1%, Backbone=ViT, multi-run=true2026.06 | 93.57 | |
| PLIFTime Step=82025.01 | 93.5 | |
| ViT-BAdaptation Strategy=Linear probing, Params=—, Pre-training Dataset=ImageNet-1K2026.05 | 93.5 | |
| LP-FTSampling Ratio=2%, Backbone=ViT2026.06 | 93.38 | |
| LP-LoRASampling Ratio=2%, Backbone=ViT2026.06 | 93.21 | |
| STBP-tdBNParam (M)=12.63, Time Step=62025.01 | 93.16 | |
| bViT-BAdaptation Strategy=Linear probing, Params=—, Pre-training Dataset=ImageNet-1K2026.05 | 93.1 | |
| STAA-SNNType=SNN Training, Architecture=ResNet-20, Timestep=12026.04 | 93.08 | |
| FSTA-SNNType=SNN Training, Architecture=ResNet-20, Timestep=12026.04 | 93.01 | |
| DA-LIFType=SNN Training, Architecture=ResNet-20, Timestep=12026.04 | 92.89 | |
| SGDNoise Level=0.4, Evaluation Protocol=linear probing, Backbone=ViT-B/162025.01 | 92.86 | |
| L3FACTSampling Ratio=1%, Backbone=ViT2026.06 | 92.76 | |
| DreamLIPZero-shot=true2025.09 | 92.7 | |
| DreamLIP2025.09 | 92.7 | |
| BEiT-B/16 + MPNetImg. encoder=BEiT-B/16, Txt. encoder=MPNet2026.05 | 92.7 | |
| Grad-MatchNoise Level=0.4, Evaluation Protocol=linear probing, Backbone=ViT-B/162025.01 | 92.62 | |
| AGRANoise Level=0.4, Evaluation Protocol=linear probing, Backbone=ViT-B/162025.01 | 92.51 |