Image Classification on DomainNet (Domain-Specific Accuracy)
98.93Accuracy (ClipArt)FedOTP
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| FedOTPTraining protocol=Prompt-based Federated Learning2024.02 | 98.93 | 98.89 | 98.42 | 84.52 | 98.64 | 87.87 | 94.55 | — | — | |
| PromptFL+FedProxTraining protocol=Prompt-based Federated Learning2024.02 | 98.45 | 96 | 98.04 | 72.32 | 96.08 | 63.51 | 87.4 | — | — | |
| CoOpTraining protocol=Local Training2024.02 | 98.32 | 98.18 | 97.7 | 83.01 | 98.21 | 82.37 | 92.95 | — | — | |
| PromptFLTraining protocol=Prompt-based Federated Learning2024.02 | 98.23 | 97.89 | 97.31 | 79.91 | 96.83 | 66.52 | 89.45 | — | — | |
| PEP-FedPT2025.10 | 95.46 | 93 | 91.79 | 71.68 | 97.67 | 86.89 | 89.15 | 59.79 | — | |
| FedVPT-D2025.10 | 94.01 | 93.45 | 91.58 | 63.29 | 96.96 | 84.56 | 87.31 | 42.49 | — | |
| Pre-IN1KModel Type=PM, Communication Rounds=5-rounds2025.03 | 93.1 | 86.4 | 88.4 | 67.8 | 94.9 | 92.7 | 87.2 | — | — | |
| WarmFedModel Type=PM, Communication Rounds=5-rounds2025.03 | 93.1 | 87.7 | 90.3 | 69.5 | 94.2 | 93.6 | 88.1 | — | — | |
| pFedPG2025.10 | 92.89 | 92.27 | 89.34 | 63.56 | 97.16 | 87.33 | 87.4 | 52.05 | — | |
| SGPT2025.10 | 92.64 | 91.54 | 89.93 | 60.62 | 96.55 | 83.55 | 85.56 | 37.34 | — | |
| HiLoRA2026.03 | 92.1 | 62.4 | 92.4 | 89.9 | 98.9 | 87.2 | 87.7 | 58.9 | — | |
| OursSelective Forget=painting, Unlearning State=After (AF)2025.12 | 91.76 | 3.51 | 85.76 | 71.95 | 96.36 | 46.03 | — | — | — | |
| OursSelective Forget=painting+real+sketch, Unlearning State=After (AF)2025.12 | 91.76 | 3.07 | 2.64 | 71.95 | 2.27 | 46.03 | — | — | — | |
| FedPR2025.10 | 91.62 | 91.16 | 86.41 | 56.2 | 96.66 | 73.72 | 82.95 | 35.18 | — | |
| WarmFedModel Type=GM, Communication Rounds=5-rounds2025.03 | 91.5 | 83 | 88.8 | 66 | 90.9 | 88.3 | 84.8 | — | — | |
| Head-Tuning2025.10 | 91.16 | 91.39 | 86.46 | 57.45 | 96.68 | 74.94 | 83.71 | 38.88 | — | |
| NSCSelective Forget=clipart, Unlearning State=Before (BF)2025.12 | 91.1 | 80.8 | 89.04 | 69.1 | 95.97 | 38.45 | — | — | — | |
| ZSL-CLIPSelective Forget=clipart, Unlearning State=Before (BF)2025.12 | 91.1 | 80.8 | 89.04 | 69.1 | 95.97 | 38.45 | — | — | — | |
| OursSelective Forget=clipart, Unlearning State=Before (BF)2025.12 | 91.1 | 80.8 | 89.04 | 69.1 | 95.97 | 38.45 | — | — | — | |
| OursSelective Forget=painting, Unlearning State=After (AF)2025.12 | 91.1 | 78.57 | 89.04 | 69.1 | 95.97 | 38.45 | — | — | — | |
| OursSelective Forget=painting+real+sketch, Unlearning State=After (AF)2025.12 | 91.1 | 76.87 | 85.3 | 69.1 | 94.14 | 38.45 | — | — | — | |
| FedVPT2025.10 | 90.84 | 92.25 | 88.24 | 58.56 | 96.78 | 77.81 | 84.23 | 37.02 | — | |
| ZSL-CLIPSelective Forget=clipart, Unlearning State=After (AF)2025.12 | 90.51 | 80.85 | 86.2 | 69 | 95.8 | 36.6 | — | — | — | |
| LipSelective Forget=clipart, Unlearning State=Before (BF)2025.12 | 90.5 | 80.3 | 88.5 | 68.6 | 95.7 | 38.1 | — | — | — | |
| P-PT2025.10 | 90.11 | 90.25 | 85.26 | 56.73 | 95.18 | 74.81 | 82.3 | 35.67 | — | |
| LipSelective Forget=clipart, Unlearning State=After (AF)2025.12 | 89.9 | 80.15 | 85.9 | 68.2 | 95.4 | 36 | — | — | — | |
| Pre-IN1KModel Type=GM, Communication Rounds=5-rounds2025.03 | 89.3 | 82 | 86.3 | 60 | 92.7 | 74.7 | 80.8 | — | — | |
| FGLModel Type=PM, Communication Rounds=5-rounds2025.03 | 89.1 | 79.1 | 86.6 | 65.4 | 92.7 | 92.8 | 84.3 | — | — | |
| FlexLoRA2026.03 | 88.7 | 56.7 | 92.6 | 77.9 | 97.5 | 87.4 | 83.9 | 52.6 | — | |
| FedIT2026.03 | 88.6 | 53.9 | 89.8 | 76.2 | 96.6 | 85.4 | 82.2 | 49.4 | — | |
| FGLModel Type=GM, Communication Rounds=5-rounds2025.03 | 88.4 | 78.2 | 86.9 | 60.3 | 89.1 | 84 | 81.2 | — | — | |
| OursSelective Forget=clipart+sketch, Unlearning State=After (AF)2025.12 | 88.16 | 80.8 | 85.19 | 69.1 | 95.97 | 38.45 | — | — | — | |
| FedDPA-F2026.03 | 87.9 | 61 | 92.2 | 87.7 | 98.8 | 84.8 | 86 | 58.3 | — | |
| NSCSelective Forget=clipart, Unlearning State=After (AF)2025.12 | 87.5 | 78.51 | 64.46 | 67.5 | 95.02 | 34.13 | — | — | — | |
| FedDPA-T2026.03 | 87.5 | 55.4 | 89.7 | 83 | 97.6 | 81.9 | 83.1 | 52.9 | — | |
| OursSelective Forget=clipart, Unlearning State=After (AF)2025.12 | 87.03 | 80.7 | 89 | 69.1 | 95.97 | 38.4 | — | — | — | |
| FDLoRA2026.03 | 86 | 56.2 | 88.2 | 82.2 | 97.4 | 80 | 82.2 | 53.5 | — | |
| PF2LoRA2026.03 | 85.8 | 56 | 88 | 82 | 97.3 | 79.9 | 82 | 53.1 | — | |
| FedSA-LoRA2026.03 | 85 | 56.8 | 86.7 | 81 | 97.1 | 79.2 | 81.4 | 53.2 | — | |
| FedAvgModel Type=PM2025.03 | 81.8 | 60.8 | 73.5 | 50.4 | 84.3 | 91.5 | 73.7 | — | — | |
| FedBNModel Type=PM2025.03 | 81.8 | 63.5 | 76.3 | 49.2 | 84.8 | 91.1 | 74.5 | — | — | |
| Centralized2025.03 | 81.2 | 60.1 | 70.8 | 50.2 | 81.1 | 92 | 72.6 | — | — | |
| Self-DistillationDistillation Scenario=Internal Data Only2026.04 | 80.57 | 46.23 | — | 20.85 | 65.38 | 23.53 | 47.31 | — | 0 | |
| FedALT2026.03 | 80.4 | 48.2 | 84.9 | 74.1 | 95.3 | 79.7 | 79.6 | 44 | — | |
| FGLCommunication Rounds=one-shot2025.03 | 80.2 | 76.2 | 73.5 | 49.9 | 85.5 | 49.7 | 69.2 | — | — | |
| Self-DistillationDistillation Scenario=Related External Data2026.04 | 80.1 | 48.15 | — | 21.53 | 68.75 | 25.28 | 48.76 | — | 1.45 | |
| DHVT-SGFLOPs=4.7, Pre-training=ImageNet-1K, Fine-tuning epochs=1002022.10 | 80.06 | 74.18 | 73.32 | — | — | — | — | — | — | |
| FedAvgModel Type=GM2025.03 | 79.8 | 52.2 | 65.8 | 43 | 79.7 | 71.7 | 65.3 | — | — | |
| WarmFedCommunication Rounds=one-shot2025.03 | 78.7 | 77 | 76.7 | 58.3 | 84.3 | 84 | 76.5 | — | — | |
| GlobalHierarchical=false2025.12 | 78.5 | — | 72.3 | — | 88.4 | 86.7 | 81.6 | — | — | |
| SE2DDistillation Scenario=Related External Data2026.04 | 78.05 | 47.76 | — | 21.98 | 68.43 | 23.81 | 48.01 | — | — | |
| DHVT-TGFLOPs=1.2, Pre-training=ImageNet-1K, Fine-tuning epochs=1002022.10 | 77.88 | 72.05 | 70.79 | — | — | — | — | — | — | |
| DKDDistillation Scenario=Internal Data Only2026.04 | 77.32 | 42.76 | — | 18.04 | 63.99 | 17.43 | 43.91 | — | 0 | |
| DaSHHierarchical=true2025.12 | 77.3 | — | 71.8 | — | 88.4 | 86.3 | 81 | — | — | |
| LSDistillation Scenario=Internal Data Only2026.04 | 77.21 | 40.95 | — | 17.21 | 60.39 | 16.42 | 42.43 | — | 0 | |
| KL-divergenceDistillation Scenario=Internal Data Only2026.04 | 77.08 | 44.17 | — | 18.16 | 67.29 | 17.19 | 44.78 | — | 0 | |
| MDSDistillation Scenario=Internal Data Only2026.04 | 76.92 | 42.14 | — | 18.08 | 63.29 | 17.44 | 43.57 | — | 0 | |
| DKDDistillation Scenario=Related External Data2026.04 | 76.53 | 44.24 | — | 18.7 | 68.29 | 16.24 | 44.8 | — | 0.89 | |
| VDPGBackbone=ViT-B/16 CLIP2024.05 | 76.3 | 67.8 | 66.6 | 49.3 | 81.5 | 17.4 | 59.8 | — | — | |
| KL-divergenceDistillation Scenario=Related External Data2026.04 | 76 | 44.77 | — | 18.89 | 70.65 | 15.53 | 45.17 | — | 0.39 | |
| MDSDistillation Scenario=Related External Data2026.04 | 75.32 | 42.28 | — | 17.53 | 67.29 | 16.12 | 43.71 | — | 0.14 | |
| LSDistillation Scenario=Related External Data2026.04 | 75.28 | 40.78 | — | 16.88 | 62.76 | 15.47 | 42.24 | — | -0.2 | |
| ResNet-50GFLOPs=3.8, Pre-training=ImageNet-1K, Fine-tuning epochs=1002022.10 | 75.22 | 66.58 | 67.77 | — | — | — | — | — | — | |
| MIROBackbone=ViT-B/16 CLIP2024.05 | 74.9 | 59.8 | 61.2 | 37.1 | 72.2 | 18.7 | 54 | — | — | |
| MUSTEvaluation Protocol=Unsupervised2024.11 | 74.83 | 61.8 | 70.31 | 56.48 | 82.88 | 19.06 | 60.89 | — | — | |
| CoOpEvaluation Protocol=Few-Shot, Labeled Samples=6 x 4 x 3452024.11 | 74.7 | 68.29 | 69.48 | 54.96 | 82.94 | 22.14 | 62.09 | — | — | |
| T2T-ViT-14GFLOPs=5.2, Pre-training=ImageNet-1K, Fine-tuning epochs=1002022.10 | 74.59 | 72.29 | 72.18 | — | — | — | — | — | — | |
| CLIP-AdapterEvaluation Protocol=Few-Shot, Labeled Samples=6 x 4 x 3452024.11 | 74.35 | 69.94 | 66.9 | 53.79 | 85.26 | 19.71 | 61.66 | — | — | |
| Tbest (upper bound)Distillation Scenario=Internal Data Only2026.04 | 74.35 | 66.66 | — | 35.83 | 78.9 | 66.19 | 64.39 | — | 0 | |
| PMTransSource Domain=rel2023.03 | 74.1 | 70 | 61.1 | 35.3 | — | 25.4 | 53.2 | — | — | |
| FedBNModel Type=GM2025.03 | 74.1 | 49.9 | 67.3 | 40.3 | 76.5 | 75.3 | 63.9 | — | — | |
| CLIP-DEvaluation Protocol=Transductive Learning2024.11 | 73.9 | 67.75 | 67.56 | 55.84 | 83.26 | 17.84 | 61.03 | — | — | |
| CLIP-DEvaluation Protocol=Unsupervised, Prompt=domain-specific templates2024.11 | 73.9 | 67.75 | 67.56 | 55.84 | 83.26 | 17.84 | 61.03 | — | — | |
| DHVT-S#Params=23.8M, GFLOPS=4.7, Training Protocol=train-from-scratch, Resolution=224x224, Epochs=3002022.10 | 73.89 | 66.08 | 68.72 | 35.11 | 83.64 | 74.38 | — | — | — | |
| DHVT-S#Params=23.8M2022.10 | 73.89 | 66.08 | 68.72 | — | — | — | — | — | — | |
| UMFC + CLIP-EEvaluation Protocol=Unsupervised2024.11 | 73.84 | 67.39 | 67.9 | 56.59 | 84.33 | 20.03 | 61.68 | — | — | |
| PMTransSource Domain=skt2023.03 | 73.8 | 62.6 | — | 33 | 77.5 | 30.9 | 55.6 | — | — | |
| Swin-TGFLOPs=4.5, Pre-training=ImageNet-1K, Fine-tuning epochs=1002022.10 | 73.51 | 72.99 | 72.37 | — | — | — | — | — | — | |
| TPTProtocol=Test-Time Adaptation2024.11 | 73.23 | 68 | 66.68 | 52.63 | 84.39 | 12.79 | 59.62 | — | — | |
| CLIP-EEvaluation Protocol=Transductive Learning2024.11 | 73.16 | 67.02 | 67.49 | 54.17 | 84.3 | 15.86 | 60.33 | — | — | |
| CLIP-EEvaluation Protocol=Unsupervised, Prompt=ensemble prompt templates2024.11 | 73.16 | 67.02 | 67.49 | 54.17 | 84.3 | 15.86 | 60.33 | — | — | |
| PromptStylerEvaluation Protocol=Transductive Learning2024.11 | 73.1 | 69.2 | 65.3 | 50.9 | 85.4 | 13.3 | 59.4 | — | — | |
| UMFCEvaluation Protocol=Unsupervised2024.11 | 73.02 | 66.36 | 66.87 | 55.3 | 83.54 | 19.67 | 60.79 | — | — | |
| UMFCEvaluation Protocol=Transductive Learning2024.11 | 73.01 | 66.89 | 67.51 | 55.44 | 83.66 | 20.14 | 61.11 | — | — | |
| UMFC-EMAProtocol=Test-Time Adaptation, Update Strategy=EMA2024.11 | 72.99 | 66.64 | 66.75 | 54.94 | 83.54 | 18.58 | 60.57 | — | — | |
| ResNeXt-50#Params=23.7M, GFLOPS=4.3, Variant=32x4d, Training Protocol=train-from-scratch, Resolution=224x224, Epochs=3002022.10 | 72.93 | 64.37 | 68.52 | 34.85 | 82.15 | 73.73 | — | — | — | |
| UMFC-MemoryProtocol=Test-Time Adaptation, Update Strategy=Memory2024.11 | 72.82 | 66.82 | 66.82 | 55.12 | 83.62 | 19.92 | 60.85 | — | — | |
| CoOpEvaluation Protocol=Few-Shot, Labeled Samples=6 x 1 x 3452024.11 | 72.73 | 66.8 | 67.27 | 53.95 | 82.53 | 19.58 | 60.48 | — | — | |
| CLIP-AdapterEvaluation Protocol=Few-Shot, Labeled Samples=6 x 1 x 3452024.11 | 72.67 | 67.84 | 65.75 | 51.69 | 84.26 | 17.82 | 60 | — | — | |
| ResNet-50#Params=24.2M, GFLOPS=3.8, Training Protocol=train-from-scratch, Resolution=224x224, Epochs=3002022.10 | 71.9 | 64.36 | 67.45 | 32.4 | 81.51 | 74.19 | — | — | — | |
| ResNet-50#Params=24.2M2022.10 | 71.9 | 64.36 | 67.45 | — | — | — | — | — | — | |
| DHVT-T#Params=6.1M, GFLOPS=1.2, Training Protocol=train-from-scratch, Resolution=224x224, Epochs=3002022.10 | 71.73 | 63.34 | 66.6 | 32.6 | 81.31 | 74.41 | — | — | — | |
| DHVT-T#Params=6.1M2022.10 | 71.73 | 63.34 | 66.6 | — | — | — | — | — | — | |
| Local-LoRA2026.03 | 71.6 | 42.8 | 85 | 74.8 | 95.4 | 79.4 | 75.5 | 37.4 | — | |
| CLIPProtocol=Test-Time Adaptation2024.11 | 71.21 | 64.61 | 64.81 | 49.47 | 82.98 | 14.23 | 57.88 | — | — | |
| CLIPEvaluation Protocol=Transductive Learning2024.11 | 71.21 | 64.61 | 64.81 | 49.47 | 82.98 | 14.23 | 57.88 | — | — | |
| CLIPEvaluation Protocol=Unsupervised, Prompt=fixed text template2024.11 | 71.21 | 64.61 | 64.81 | 49.47 | 82.98 | 14.23 | 57.88 | — | — | |
| SSRTSource Domain=skt2023.03 | 70.6 | 62.2 | — | 32.8 | 73.2 | 21.7 | 52.1 | — | — | |
| FreeSelHierarchical=false2025.12 | 70.1 | — | 67.2 | — | 85.6 | 81.7 | 77.7 | — | — | |
| RamenBackbone=ViT-B/322026.04 | 70.1 | 65.8 | 61.5 | 45.2 | 82.4 | 17.6 | 57.1 | — | — |