Image Classification on DTD (test)
95.8AccuracyMultitask
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MultitaskBackbone=ViT-B/32, #Params=x1, Merging Strategy=N/A2026.05 | 95.8 | — | — | — | |
| SDFedBackbone=ViT-B162026.02 | 93.56 | — | — | — | |
| UOPPBackbone=ViT-B162026.02 | 93.09 | — | — | — | |
| GPT-FLBackbone=ViT-B162026.02 | 92.86 | — | — | — | |
| FedOTPBackbone=ViT-B162026.02 | 89.84 | — | — | — | |
| SDFedBackbone=ResNet502026.02 | 89.45 | — | — | — | |
| UOPPBackbone=ResNet502026.02 | 87.56 | — | — | — | |
| GPT-FLBackbone=ResNet502026.02 | 86.42 | — | — | — | |
| WiSE-FTalpha=optimal2021.09 | 85.2 | — | — | — | |
| WiSE-FTalpha=0.52021.09 | 84.6 | — | — | — | |
| PowderBackbone=ViT-B162026.02 | 83.23 | — | — | — | |
| CoOpBackbone=ViT-B162026.02 | 82.91 | — | — | — | |
| C3ABackbone=ViT-Large, Config=b=1024/16, # Params=0.79M2024.07 | 82.62 | — | — | — | |
| FedOTPBackbone=ResNet502026.02 | 82.15 | — | — | — | |
| Standard fine-tuning2021.09 | 81.9 | — | — | — | |
| Linear ProbeInterpretability=false, Evaluation Setting=Fully-supervised, Backbone=CLIP ViT-L/142026.01 | 81.68 | — | — | — | |
| PCBM-ReDInterpretability=true, Evaluation Setting=Fully-supervised, Backbone=CLIP ViT-L/142026.01 | 81.44 | — | — | — | |
| C3ABackbone=ViT-Base, Config=b=768/12, # Params=0.22M2024.07 | 80.57 | — | — | — | |
| FullBackbone=ViT-Large, # Params=303M2024.07 | 80.18 | — | — | — | |
| FullBackbone=ViT-Base, # Params=85.8M2024.07 | 80.11 | — | — | — | |
| LoRABackbone=ViT-Large, Rank=r=16, # Params=1.57M2024.07 | 80.09 | — | — | — | |
| Ensemblemode=ensemble2021.07 | 79.79 | — | — | — | |
| Fine-tuned (STL)Backbone=CLIP-ViT-B/32, Evaluation Protocol=Fine-tuned (STL)2026.05 | 79.7 | — | — | — | |
| IndividualsBackbone=ViT-B/32, #Params=x8, Merging Strategy=N/A2026.05 | 79.4 | — | — | — | |
| Non-binary Transfer Learning (Optimal)reinitialized layers=2 new layers, learning rate=0.002, L2SP alpha=0.001, L2SP beta=0.001, L2SP layers=14, regularization=L2SP2021.07 | 78.9 | — | — | — | |
| State of the art2021.07 | 78.9 | — | — | — | |
| LoRABackbone=ViT-Base, Rank=r=16, # Params=0.59M2024.07 | 78.56 | — | — | — | |
| V2C-CBMInterpretability=true, Evaluation Setting=Fully-supervised, Backbone=CLIP ViT-L/142026.01 | 78.49 | — | — | — | |
| L2SP (Default)learning rate=0.002, regularization=L2SP2021.07 | 78.32 | — | — | — | |
| SMILEBackbone=ViT-B/32, #Params=x3.07, Merging Strategy=Dynamic Merging2026.05 | 78.1 | — | — | — | |
| Traditional MTLBackbone=CLIP-ViT-B/32, Evaluation Protocol=Traditional MTL2026.05 | 77.7 | — | — | — | |
| LaBoInterpretability=true, Evaluation Setting=Fully-supervised, Backbone=CLIP ViT-L/142026.01 | 77.3 | — | — | — | |
| L2 (Default)learning rate=0.002, regularization=L22021.07 | 77.3 | — | — | — | |
| WUDI-Merging w/ FEATCALBackbone=CLIP-ViT-B/32, Upstream Merger=WUDI-Merging, Refinement=FEATCAL2026.05 | 77.1 | — | — | — | |
| WEMoEBackbone=ViT-B/32, #Params=x6.27, Merging Strategy=Dynamic Merging2026.05 | 76.6 | — | — | — | |
| PowderBackbone=ResNet502026.02 | 76.54 | — | — | — | |
| TransBoostEvaluation Protocol=Transductive, Resolution=224, Backbone=ResNet50, Classes=47, Train (+Validation) size=3,760, Test size=1,8802022.05 | 76.49 | — | 1.7 | — | |
| AdaMerging w/ FEATCALBackbone=CLIP-ViT-B/32, Upstream Merger=AdaMerging, Refinement=FEATCAL2026.05 | 76.4 | — | — | — | |
| DiDi-Merging-MBackbone=ViT-B/32, #Params=x1.4, Merging Strategy=Dynamic Merging2026.05 | 76.2 | — | — | — | |
| Res-CBMInterpretability=true, Evaluation Setting=Fully-supervised, Backbone=CLIP ViT-L/142026.01 | 75.77 | — | — | — | |
| SupConEvaluation Protocol=Full-data finetuning, Pre-trained Dataset=ImageNet-1K, Pre-training Epochs=3002021.12 | 75.43 | — | — | — | |
| FREE-MergingBackbone=ViT-B/32, #Params=x2.16, Merging Strategy=Dynamic Merging2026.05 | 75.4 | — | — | — | |
| TSV-CBackbone=ViT-B/32, #Params=x2.08, Merging Strategy=Dynamic Merging2026.05 | 75.3 | — | — | — | |
| CoOpBackbone=ResNet502026.02 | 75.12 | — | — | — | |
| Standard InductiveEvaluation Protocol=Inductive, Resolution=224, Backbone=ResNet50, Classes=47, Train (+Validation) size=3,760, Test size=1,8802022.05 | 74.79 | — | — | — | |
| TALL-Mask+TIESBackbone=ViT-B/32, #Params=x2.5, Merging Strategy=Dynamic Merging2026.05 | 74.4 | — | — | — | |
| SL-MLPEvaluation Protocol=Full-data finetuning, Pre-trained Dataset=ImageNet-1K, Pre-training Epochs=3002021.12 | 74.26 | — | — | — | |
| DiDi-Merging-SBackbone=ViT-B/32, #Params=x1.24, Merging Strategy=Dynamic Merging2026.05 | 74.1 | — | — | — | |
| SupCon w/o MLPEvaluation Protocol=Full-data finetuning, Pre-trained Dataset=ImageNet-1K, Pre-training Epochs=3002021.12 | 74.04 | — | — | — | |
| Twin-MergingBackbone=ViT-B/32, #Params=x2.25, Merging Strategy=Dynamic Merging2026.05 | 74 | — | — | — | |
| AdaMerging w/ ProbSurgeryBackbone=CLIP-ViT-B/32, Upstream Merger=AdaMerging, Refinement=ProbSurgery2026.05 | 73.8 | — | — | — | |
| SLEvaluation Protocol=Full-data finetuning, Pre-trained Dataset=ImageNet-1K, Pre-training Epochs=3002021.12 | 73.78 | — | — | — | |
| MMCLBackbone=ResNet-50, Pre-training=ImageNet-1K, Evaluation Setting=Fine-tuning2021.12 | 73.51 | — | — | — | |
| HeadBackbone=ViT-Large2024.07 | 73.33 | — | — | — | |
| SupervisedBackbone=ResNet-50, Evaluation Setting=Fine-tuning2021.12 | 73.3 | — | — | — | |
| AdaMerging w/ SurgeryBackbone=CLIP-ViT-B/32, Upstream Merger=AdaMerging, Refinement=Surgery2026.05 | 73.3 | — | — | — | |
| WUDI-Merging w/ ProbSurgeryBackbone=CLIP-ViT-B/32, Upstream Merger=WUDI-Merging, Refinement=ProbSurgery2026.05 | 73.3 | — | — | — | |
| WUDI-Merging w/ SurgeryBackbone=CLIP-ViT-B/32, Upstream Merger=WUDI-Merging, Refinement=Surgery2026.05 | 72.7 | — | — | — | |
| EMR-MergingBackbone=ViT-B/32, #Params=x4, Merging Strategy=Dynamic Merging2026.05 | 72.1 | — | — | — | |
| SaSPAResolution=448x4482024.06 | 72 | — | — | — | |
| PCL-v2Backbone=ResNet-50, Pre-training=ImageNet-1K, Evaluation Setting=Fine-tuning2021.12 | 71.76 | — | — | — | |
| CLIP-B32 (LOES)Model=CLIP-B32, Selection Strategy=LOES2026.05 | 70.93 | — | — | — | |
| MOCHIBackbone=ResNet-50, Pre-training=ImageNet-1K, Evaluation Setting=Fine-tuning, Public Checkpoint=true2021.12 | 70.85 | — | — | — | |
| ProDAshots=162022.05 | 70.1 | — | — | — | |
| PCL-v1Backbone=ResNet-50, Pre-training=ImageNet-1K, Evaluation Setting=Fine-tuning2021.12 | 70 | — | — | — | |
| AiRPrompting Strategy=Text Prompt, Learning Paradigm=TRZSL2025.04 | 69.9 | — | — | — | |
| Task Arithmetic w/ FEATCALBackbone=CLIP-ViT-B/32, Upstream Merger=Task Arithmetic, Refinement=FEATCAL2026.05 | 69.8 | — | — | — | |
| HeadBackbone=ViT-Base2024.07 | 69.77 | — | — | — | |
| WUDI-MergingBackbone=ViT-B/32, #Params=x1, Merging Strategy=Static Merging2026.05 | 69.7 | — | — | — | |
| Best AugResolution=448x4482024.06 | 69.6 | — | — | — | |
| MoCo-v2Backbone=ResNet-50, Pre-training=ImageNet-1K, Evaluation Setting=Fine-tuning, Public Checkpoint=true2021.12 | 69.41 | — | — | — | |
| TSV-MergingBackbone=ViT-B/32, #Params=x1, Merging Strategy=Static Merging2026.05 | 69.2 | — | — | — | |
| Simple Averaging w/ ProbSurgeryBackbone=CLIP-ViT-B/32, Upstream Merger=Simple Averaging, Refinement=ProbSurgery2026.05 | 69.1 | — | — | — | |
| DINOv3-S/16 (LOES)Model=DINOv3-S/16, Selection Strategy=LOES2026.05 | 68.24 | — | — | — | |
| DINOv2-S (LOES)Model=DINOv2-S, Selection Strategy=LOES2026.05 | 68.1 | — | — | — | |
| CPLYear=2024, Prompting Strategy=Text Prompt, Learning Paradigm=TRZSL2025.04 | 68 | — | — | — | |
| DINO.txt#param=304M, Evaluation Protocol=Zero-shot2026.04 | 67.5 | — | — | — | |
| Simple Averaging w/ FEATCALBackbone=CLIP-ViT-B/32, Upstream Merger=Simple Averaging, Refinement=FEATCAL2026.05 | 67.4 | — | — | — | |
| ISO-CTSBackbone=ViT-B/32, #Params=x1, Merging Strategy=Static Merging2026.05 | 67 | — | — | — | |
| DeiT-B/16 (LOES)Model=DeiT-B/16, Selection Strategy=LOES2026.05 | 66.95 | — | — | — | |
| AdaMergingBackbone=CLIP-ViT-B/32, Upstream Merger=AdaMerging2026.05 | 66.9 | — | — | — | |
| WUDI-MergingBackbone=CLIP-ViT-B/32, Upstream Merger=WUDI-Merging2026.05 | 66.7 | — | — | — | |
| AiRPrompting Strategy=Visual Prompt, Learning Paradigm=TRZSL2025.04 | 66.4 | — | — | — | |
| MoCoBackbone=ResNet-50, Pre-training=ImageNet-1K, Evaluation Setting=Fine-tuning2021.12 | 65.37 | — | — | — | |
| GRIPYear=2023, Prompting Strategy=Text Prompt, Learning Paradigm=TRZSL2025.04 | 65.3 | — | — | — | |
| CPLYear=2024, Prompting Strategy=Visual Prompt, Learning Paradigm=TRZSL2025.04 | 65.3 | — | — | — | |
| US3LBackbone=ResNet-18, Width=1.0x, Params=11.69M, MACS=1.82G, Evaluation Protocol=linear evaluation2023.03 | 65.2 | — | — | — | |
| CoOpshots=162022.05 | 65 | — | — | — | |
| ProDAshots=82022.05 | 65 | — | — | — | |
| BYOLBackbone=ResNet-18, Width=1.0x, Params=11.69M, MACS=1.82G, Evaluation Protocol=linear evaluation2023.03 | 64.8 | — | — | — | |
| Simple Averaging w/ SurgeryBackbone=CLIP-ViT-B/32, Upstream Merger=Simple Averaging, Refinement=Surgery2026.05 | 64.7 | — | — | — | |
| Linear Probe CLIPshots=162022.05 | 64.3 | — | — | — | |
| InsDisBackbone=ResNet-50, Pre-training=ImageNet-1K, Evaluation Setting=Fine-tuning2021.12 | 63.99 | — | — | — | |
| Task Arithmetic w/ ProbSurgeryBackbone=CLIP-ViT-B/32, Upstream Merger=Task Arithmetic, Refinement=ProbSurgery2026.05 | 63.8 | — | — | — | |
| BYOLBackbone=ResNet-18, Width=0.75x, Params=6.68M, MACS=1.05G, Evaluation Protocol=linear evaluation2023.03 | 63.7 | — | — | — | |
| US3LBackbone=ResNet-18, Width=0.75x, Params=6.68M, MACS=1.05G, Evaluation Protocol=linear evaluation2023.03 | 63.7 | — | — | — | |
| ViT-IN21k-B/16 (LOES)Model=ViT-IN21k-B/16, Selection Strategy=LOES2026.05 | 63.68 | — | — | — | |
| CLIP-B32 (Last)Model=CLIP-B32, Selection Strategy=Last Layer2026.05 | 62.96 | — | — | — | |
| PCBM-ReD + CuPLInterpretability=true, Evaluation Setting=Zero-shot, Backbone=CLIP ViT-L/142026.01 | 62.77 | — | — | — | |
| GRIPYear=2023, Prompting Strategy=Visual Prompt, Learning Paradigm=TRZSL2025.04 | 62.7 | — | — | — |