Image Classification on DTD (Absolute, Normalized, Average Accuracy)
86.12AccuracyθB fine-tune
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| θB fine-tuneTarget Backbone=ViT-L/14, Target Pre-training=Datacomp-XL, Evaluation Protocol=fine-tune2026.05 | 86.12 | — | — | — | |
| theta_A fine-tuneN=–2026.05 | 84.25 | — | — | — | |
| θA fine-tuneSource Backbone=ViT-B/16, Source Pre-training=LAION-2B, Evaluation Protocol=fine-tune2026.05 | 83.46 | — | — | — | |
| theta_B fine-tuneN=–2026.05 | 82.55 | — | — | — | |
| SCANShot=162025.05 | 78.53 | — | — | — | |
| SCANShot=82025.05 | 75.28 | — | — | — | |
| EBMmode=Fine-tuning2025.10 | 73.3 | — | — | — | |
| SCANShot=42025.05 | 72.91 | — | — | — | |
| PromptSRCmode=Fine-tuning2025.10 | 72.7 | — | — | — | |
| BiCoSource Backbone=ViT-B/16, Source Pre-training=LAION-2B, Target Backbone=ViT-L/14, Target Pre-training=Datacomp-XL, N=100, Evaluation Protocol=task-vector transfer2026.05 | 71.54 | — | — | — | |
| PLOTmode=Fine-tuning2025.10 | 71.4 | — | — | — | |
| MaPLemode=Fine-tuning2025.10 | 71.3 | — | — | — | |
| SimNLShot=162025.05 | 70.83 | — | — | — | |
| CoOpmode=Fine-tuning2025.10 | 69.9 | — | — | — | |
| SimNLShot=82025.05 | 67.78 | — | — | — | |
| Tip-Adapter-FShot=162025.05 | 67.33 | — | — | — | |
| TaskResShot=162025.05 | 67.13 | — | — | — | |
| θB zero-shotTarget Backbone=ViT-L/14, Target Pre-training=Datacomp-XL, Evaluation Protocol=zero-shot2026.05 | 66.96 | — | — | — | |
| θB + τA paddedSource Backbone=ViT-B/16, Source Pre-training=LAION-2B, Target Backbone=ViT-L/14, Target Pre-training=Datacomp-XL, Evaluation Protocol=task-vector transfer2026.05 | 66.75 | — | — | — | |
| CLIP-AdapterShot=162025.05 | 66.14 | — | — | — | |
| SimNLShot=42025.05 | 66.01 | — | — | — | |
| SCANShot=22025.05 | 65.47 | — | — | — | |
| TaskResShot=82025.05 | 64.77 | — | — | — | |
| CoOpShot=162025.05 | 63.11 | — | — | — | |
| CoCoOpmode=Fine-tuning2025.10 | 63 | — | — | — | |
| Tip-Adapter-FShot=82025.05 | 62.67 | — | — | — | |
| BiCoN=1002026.05 | 62.66 | — | — | — | |
| TaskResShot=42025.05 | 60.7 | — | — | — | |
| CLIP-AdapterShot=82025.05 | 60.7 | — | — | — | |
| BiCoN=502026.05 | 60.43 | — | — | — | |
| CoOpShot=82025.05 | 59.88 | — | — | — | |
| SimNLShot=22025.05 | 59.17 | — | — | — | |
| GPUASamples per class=full training set2026.06 | 58.5 | — | — | — | |
| SCANShot=12025.05 | 58.43 | — | — | — | |
| BiCoN=202026.05 | 58.35 | — | — | — | |
| THESEUSSource Backbone=ViT-B/16, Source Pre-training=LAION-2B, Target Backbone=ViT-L/14, Target Pre-training=Datacomp-XL, N=100, Evaluation Protocol=task-vector transfer2026.05 | 58.24 | — | — | — | |
| COSMICInference-time adaptation=true2026.06 | 58.2 | — | — | — | |
| Tip-Adapter-FShot=42025.05 | 57.78 | — | — | — | |
| CoCoOpShot=162025.05 | 57.53 | — | — | — | |
| CLIP-AdapterShot=42025.05 | 57.02 | — | — | — | |
| BiCoN=102026.05 | 57.02 | — | — | — | |
| GPUA*Samples per class=162026.06 | 56.7 | — | — | — | |
| THESEUSN=202026.05 | 56.17 | — | — | — | |
| THESEUSN=1002026.05 | 55.85 | — | — | — | |
| DMN2026.06 | 55.8 | — | — | — | |
| THESEUSN=502026.05 | 55.74 | — | — | — | |
| THESEUSN=102026.05 | 55.53 | — | — | — | |
| theta_B zero-shotN=–2026.05 | 55.42 | — | — | — | |
| TaskResShot=22025.05 | 55.13 | — | — | — | |
| theta_B + tau_A^paddedN=–2026.05 | 54.79 | — | — | — | |
| DPE2026.06 | 54.2 | — | — | — | |
| Tip-Adapter-FShot=22025.05 | 54 | — | — | — | |
| CoOpShot=42025.05 | 53.38 | — | — | — | |
| SimNLShot=12025.05 | 53.13 | — | — | — | |
| CoCoOpShot=82025.05 | 52.21 | — | — | — | |
| CLIP-AdapterShot=22025.05 | 51.81 | — | — | — | |
| Tip-Adapter-FShot=12025.05 | 50.31 | — | — | — | |
| TaskResShot=12025.05 | 50.2 | — | — | — | |
| TIPPLEInference-time adaptation=true2026.06 | 49.2 | — | — | — | |
| ZLaP2026.06 | 48.6 | — | — | — | |
| StatA2026.06 | 48.4 | — | — | — | |
| CoCoOpShot=42025.05 | 47.9 | — | — | — | |
| TDA2026.06 | 47.4 | — | — | — | |
| PromptSRCBackbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 46.87 | — | — | — | |
| CLIP-AdapterShot=12025.05 | 46.66 | — | — | — | |
| MMABackbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 46.57 | — | — | — | |
| MaPLeBackbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 46.49 | — | — | — | |
| MAIL++Backbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 46.37 | — | — | — | |
| PromptSRCSource Dataset=ImageNet, Evaluation Protocol=Cross-dataset transfer (zero-shot)2025.11 | 46.27 | — | — | — | |
| CoCoOpShot=22025.05 | 46.2 | — | — | — | |
| BIP-DBackbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 46 | — | — | — | |
| MMRLBackbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 45.9 | — | — | — | |
| ZERO2026.06 | 45.9 | — | — | — | |
| MTA2026.06 | 45.9 | — | — | — | |
| CoOpOpBackbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 45.73 | — | — | — | |
| MAILBackbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 45.47 | — | — | — | |
| MaPLeSource Dataset=ImageNet, Evaluation Protocol=Cross-dataset transfer (zero-shot)2025.11 | 45.39 | — | — | — | |
| CoCoOpShot=12025.05 | 45.14 | — | — | — | |
| CoOpShot=22025.05 | 45.04 | — | — | — | |
| DeKgBackbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 44.47 | — | — | — | |
| TCPBackbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 44.35 | — | — | — | |
| CoOpShot=12025.05 | 44.13 | — | — | — | |
| CLIPmode=Fine-tuning2025.10 | 44.1 | — | — | — | |
| CLIP2026.06 | 43.5 | — | — | — | |
| CoOpBackbone=ViT-B/16, Training Shots=16-shot, Training Source=ImageNet2026.05 | 41.92 | — | — | — | |
| Co-CoOpSource Dataset=ImageNet, Evaluation Protocol=Cross-dataset transfer (zero-shot)2025.11 | 41.73 | — | — | — | |
| CLIPShot=02025.05 | 40.33 | — | — | — | |
| AdaMergingBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=Vanilla Merging2026.03 | — | — | 73.5 | 70 | |
| DARE-TIESBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=Vanilla Merging2026.03 | — | — | 74.5 | 63.1 | |
| EMR-MergingBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=Vanilla Merging2026.03 | — | — | 73.9 | 64 | |
| FinetunedBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=Baseline2026.03 | — | 58.3 | — | 84.1 | |
| FR-MergingBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=Vanilla Merging2026.03 | — | — | 71 | 57.6 | |
| GCENoise Type=Asymmetric Label Noise, Noise Rate=75.0%2026.04 | — | 18.23 | — | — | |
| Iso-CBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=Vanilla Merging2026.03 | — | — | 82.1 | 73.6 | |
| Iso-CTSBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=Vanilla Merging2026.03 | — | — | 82.3 | 73.5 | |
| KnOTS-DARE-TIESBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=LoRA-aware Merging2026.03 | — | — | 75.9 | 63.9 | |
| KnOTS-TIESBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=LoRA-aware Merging2026.03 | — | — | 73.7 | 68 | |
| LinearBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=LoRA-aware Merging2026.03 | — | — | 74.5 | 63.7 | |
| LoRA-LEGOBackbone=ViT-B/32, Pre-trained=ImageNet-21k, Fine-tuning Protocol=rank-16 LoRA, Merging Category=LoRA-aware Merging2026.03 | — | — | 73 | 62.9 | |
| NLPromptNoise Type=Symmetric Label Noise, Noise Rate=37.5%2026.04 | — | 59.17 | — | — |