Image Classification on RESISC45
98.9AccuracyIndividual
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| IndividualBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=12024.05 | 98.9 | — | — | — | |
| θA fine-tuneNumber of calibration samples (N)=-2026.05 | 97.58 | — | — | — | |
| fine-tunedBackbone=ViT-L/142026.02 | 97.4 | — | — | — | |
| RandLoRA (r = 10)Backbone=CLIP-ViT-L/14, Params (M)=12.782026.04 | 97.4 | — | — | — | |
| PaRaMSModel Status=Protected Standalone (θ̂_def)2025.11 | 97.37 | — | — | — | |
| DOFABackbone=ViT-L2026.03 | 97.3 | — | — | — | |
| Full Fine-TuningBackbone=CLIP-ViT-L/14, Params (M)=303.282026.04 | 97.3 | — | — | — | |
| MergeGuardModel Status=Protected Standalone (θ̂_def)2025.11 | 97.25 | — | — | — | |
| Full Fine-TuningBackbone=DINOv2-ViT-B/14, Params (M)=86.732026.04 | 97.2 | — | — | — | |
| LoRA (r = 32)Backbone=CLIP-ViT-L/14, Params (M)=12.582026.04 | 97.2 | — | — | — | |
| theta_B fine-tuneBackbone=ViT-L/142025.10 | 97.06 | — | — | — | |
| GiVA (r = 32, SecSV-B)Backbone=CLIP-ViT-L/14, Params (M)=0.202026.04 | 97 | — | — | — | |
| GiVA (r = 32, Rand-B)Backbone=CLIP-ViT-L/14, Params (M)=0.202026.04 | 97 | — | — | — | |
| RandLoRA (r = 10)Backbone=DINOv2-ViT-B/14, Params (M)=3.592026.04 | 96.9 | — | — | — | |
| VeRA (r = 256)Backbone=CLIP-ViT-L/14, Params (M)=0.232026.04 | 96.9 | — | — | — | |
| GiVA (r = 32, TopSV-B)Backbone=CLIP-ViT-L/14, Params (M)=0.202026.04 | 96.9 | — | — | — | |
| GeoMELTBackbone=BEiT-3, zero-shot=true2025.12 | 96.81 | — | — | — | |
| theta_B fine-tuneModel=ViT-L/14, Mode=fine-tune2026.02 | 96.73 | — | — | — | |
| theta_A fine-tuneN=–2026.05 | 96.71 | — | — | — | |
| θB fine-tuneNumber of calibration samples (N)=-2026.05 | 96.71 | — | — | — | |
| fine-tunedBackbone=ViT-B/162026.02 | 96.6 | — | — | — | |
| LoRA (r = 32)Backbone=DINOv2-ViT-B/14, Params (M)=4.722026.04 | 96.6 | — | — | — | |
| GeoSANEBackbone=ViT-L2026.03 | 96.5 | — | — | — | |
| theta_B fine-tuneN=–2026.05 | 96.46 | — | — | — | |
| FFBackbone=ViT-Large, Method=FF, # Params=303.3M2026.05 | 96.43 | — | — | — | |
| ISO-CTS/LARVBackbone=ViT-L/142026.02 | 96.4 | — | — | 0.8 | |
| FTBackbone=ViT-Large, Memory=4629.8 MB2026.05 | 96.4 | — | — | — | |
| GiVA (r = 32, TopSV-B)Backbone=DINOv2-ViT-B/14, Params (M)=0.082026.04 | 96.2 | — | — | — | |
| FFBackbone=ViT-Base, Method=FF, # Params=85.8M2026.05 | 96.13 | — | — | — | |
| IndividualBackbone=ViT-B/322026.04 | 96.11 | — | — | — | |
| Fine-Tuned*Note=Upper-bound2025.12 | 96.1 | — | — | — | |
| CERSABackbone=ViT-Large, Memory=1232.3 MB2026.05 | 96.1 | — | — | — | |
| FTBackbone=ViT-Base2026.05 | 96.1 | — | — | — | |
| IndividualBackbone=CLIP ViT-B/32, Merging Category=Non-merging2026.05 | 96.1 | — | — | — | |
| VeRA (r = 256)Backbone=DINOv2-ViT-B/14, Params (M)=0.092026.04 | 96 | — | — | — | |
| Auto-FlexSwitch+CKA (λ = 3)Type=Dynamic, Storage (MB)=47.26, Backbone=ViT-B/322026.04 | 95.92 | — | — | — | |
| ISO-C/LARVBackbone=ViT-L/142026.02 | 95.9 | — | — | 0.5 | |
| GiVA (r = 32, SecSV-B)Backbone=DINOv2-ViT-B/14, Params (M)=0.082026.04 | 95.9 | — | — | — | |
| GiVA (r = 32, Rand-B)Backbone=DINOv2-ViT-B/14, Params (M)=0.082026.04 | 95.8 | — | — | — | |
| Auto-FlexSwitch+KL (λ = 0.3)Type=Dynamic, Storage (MB)=41.51, Backbone=ViT-B/322026.04 | 95.71 | — | — | — | |
| IndividualBackbone=ConvNeXt, Storage (MB)=–2026.04 | 95.71 | — | — | — | |
| Scale-MAEBackbone=ViT-L2026.03 | 95.7 | — | — | — | |
| RingMoBackbone=Swin-B2026.03 | 95.7 | — | — | — | |
| PiSSABackbone=ViT-Large, Memory=1229.9 MB2026.05 | 95.7 | — | — | — | |
| theta_B fine-tuneBackbone=ViT-B/162025.10 | 95.66 | — | — | — | |
| theta_A fine-tuneModel=laion2b, Protocol=fine-tune2026.02 | 95.65 | — | — | — | |
| CERSABackbone=ViT-Base, α=0.95, β=0.952026.05 | 95.6 | — | — | — | |
| Auto-FlexSwitch+MSE (λ = 0.05)Type=Dynamic, Storage (MB)=50.72, Backbone=ViT-B/322026.04 | 95.57 | — | — | — | |
| PiSSABackbone=ViT-Base, Rank=322026.05 | 95.5 | — | — | — | |
| FinetuningBackbone=ViT-B/322026.07 | 95.44 | — | — | — | |
| θBmode=fine-tune2026.02 | 95.4 | — | — | — | |
| θBMode=fine-tune2026.02 | 95.4 | — | — | — | |
| theta_A fine-tuneModel=ViT-B/16, Mode=fine-tune2026.02 | 95.4 | — | — | — | |
| SVFTBackbone=ViT-Large, Memory=1355.8 MB2026.05 | 95.4 | — | — | — | |
| Uni-LoRABackbone=ViT-Large, Method=Uni-LoRA, # Params=144K2026.05 | 95.24 | — | — | — | |
| SVFitBackbone=ViT-Large, Memory=1351.5 MB2026.05 | 95.2 | — | — | — | |
| FourierFTBackbone=ViT-Large, Method=FourierFT, # Params=480K2026.05 | 95.2 | — | — | — | |
| Twin-MergingType=Dynamic, Storage (MB)=711.10, Backbone=ViT-B/322026.04 | 95.13 | — | — | — | |
| fine-tunedBackbone=ViT-B/322026.02 | 95.1 | — | — | — | |
| GPartBackbone=ViT-Large, Method=GPart (Ours), # Params=144K2026.05 | 95.09 | — | — | — | |
| CLIPArchitecture=ViT-B/16, Pre-training Data=LAION-400M, Protocol=5-way 5-shot2023.05 | 95 | — | — | — | |
| LaCLIPArchitecture=ViT-B/16, Pre-training Data=LAION-400M, Protocol=5-way 5-shot2023.05 | 95 | — | — | — | |
| SatMAEBackbone=ViT-L2026.03 | 94.8 | — | — | — | |
| Auto-FlexSwitch+MSE (λ = 0.5)Type=Dynamic, Backbone=ConvNeXt, Storage (MB)=8.252026.04 | 94.79 | — | — | — | |
| LoRABackbone=ViT-Large, Memory=1229.9 MB2026.05 | 94.7 | — | — | — | |
| Auto-SwitchType=Dynamic, Storage (MB)=185.28, Backbone=ViT-B/322026.04 | 94.48 | — | — | — | |
| Auto-FlexSwitch+KL (λ = 3)Type=Dynamic, Backbone=ConvNeXt, Storage (MB)=13.652026.04 | 94.44 | — | — | — | |
| Auto-FlexSwitch+CKA (λ = 30)Type=Dynamic, Backbone=ConvNeXt, Storage (MB)=9.412026.04 | 94.43 | — | — | — | |
| TSV-M/LARVBackbone=ViT-L/142026.02 | 94.4 | — | — | 0.9 | |
| FFT MoE# Params (%)=770, Backbone=CLIP ViT-B/322026.04 | 94.38 | — | — | — | |
| CLIPArchitecture=ViT-B/32, Pre-training Data=LAION-400M, Protocol=5-way 5-shot2023.05 | 94.3 | — | — | — | |
| GPartBackbone=ViT-Base, Method=GPart (Ours), # Params=72K2026.05 | 94.29 | — | — | — | |
| FourierFTBackbone=ViT-Base, Method=FourierFT, # Params=239K2026.05 | 94.26 | — | — | — | |
| LaCLIPArchitecture=ViT-B/32, Pre-training Data=LAION-400M, Protocol=5-way 5-shot2023.05 | 94.2 | — | — | — | |
| MoW-MergingType=Dynamic, Backbone=ConvNeXt, Storage (MB)=3014.752026.04 | 94.11 | — | — | — | |
| Uni-LoRABackbone=ViT-Base, Method=Uni-LoRA, # Params=72K2026.05 | 94.08 | — | — | — | |
| Traditional MTLBackbone=CLIP ViT-B/32, Merging Category=Non-merging2026.05 | 93.9 | — | — | — | |
| FourierFTBackbone=ViT-Large, Method=FourierFT, # Params=144K2026.05 | 93.86 | — | — | — | |
| FFT# Params (%)=100, Backbone=CLIP ViT-B/322026.04 | 93.65 | — | — | — | |
| FourierMoE# Params (%)=0.42, Backbone=CLIP ViT-B/322026.04 | 93.65 | — | — | — | |
| MoW-MergingType=Dynamic, Storage (MB)=2171.89, Backbone=ViT-B/322026.04 | 93.59 | — | — | — | |
| EMR-MergingType=Dynamic, Storage (MB)=540.19, Backbone=ViT-B/322026.04 | 93.49 | — | — | — | |
| GOAT# Params (%)=2.24, Backbone=CLIP ViT-B/322026.04 | 93.46 | — | — | — | |
| Traditional MTLBackbone=ViT-B/322026.04 | 93.43 | — | — | — | |
| ISO-CTS/LARVBackbone=ViT-B/162026.02 | 93.4 | — | — | 0.8 | |
| DREAMWay=5, Shot=52026.03 | 93.4 | — | — | — | |
| MDA AMBackbone=ViT-B/32, Rotation Alignment=true2025.11 | 93.3 | — | — | — | |
| ISO-C/LARVBackbone=ViT-B/162026.02 | 93.2 | — | — | 0.9 | |
| MDA AMBackbone=ViT-B/32, Rotation Alignment=false2025.11 | 93.1 | — | — | — | |
| EMR-MERGINGBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=302024.05 | 93.01 | — | — | — | |
| SVFitBackbone=ViT-Base, Rank=7682026.05 | 93 | — | — | — | |
| HydraLoRA# Params (%)=1.58, Backbone=CLIP ViT-B/322026.04 | 92.93 | — | — | — | |
| CLIP+Pre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.07 | 92.9 | — | — | — | |
| LoRA (rank32)# Params (%)=5.98, Backbone=CLIP ViT-B/322026.04 | 92.87 | — | — | — | |
| LaSLIPArchitecture=ViT-B/16, Pre-training Data=CC12M, Protocol=5-way 5-shot2023.05 | 92.8 | — | — | — | |
| UNICOMPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.07 | 92.7 | — | — | — | |
| LoRABackbone=ViT-Base, Rank=322026.05 | 92.7 | — | — | — | |
| theta_B fine-tuneModel=laion400m, Protocol=fine-tune2026.02 | 92.59 | — | — | — | |
| MoLoRA# Params (%)=2.24, Backbone=CLIP ViT-B/322026.04 | 92.58 | — | — | — | |
| CLIPPre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.07 | 92.5 | — | — | — |