Image Classification on ImageNet Source to 10 Fine-Grained Targets (test)
96.96Caltech101 AccuracyFGA
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FGABackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 96.96 | 91.28 | 68.93 | 72.11 | 26.97 | 69.29 | 49.76 | 47.58 | 84.95 | 68.17 | 67.6 | — | |
| LPT2024.10 | 95 | 90.87 | 66.5 | 73.4 | 25.67 | 68 | 43.95 | 54.13 | 86.53 | 69.27 | 67.33 | 71.77 | |
| EvoPromptTraining Source=ImageNet, Evaluation Mode=Cross-Dataset Transfer2026.03 | 94.9 | 90.4 | 65.8 | 72.5 | 24.2 | 67.4 | 46.92 | 49.4 | 87.4 | 69.3 | 66.82 | 71.63 | |
| HeBAYear=Ours, Training Shots=16 shots, Source Training Dataset=ImageNet2026.03 | 94.81 | 92.2 | 65.41 | 73.04 | 27.09 | 68.22 | 50.71 | 58.99 | 86.13 | 70.45 | 68.71 | 71.5 | |
| LwEIBYear=2025, Training Shots=16 shots, Source Training Dataset=ImageNet2026.03 | 94.51 | 92.5 | 66.58 | 73.03 | 27.7 | 69.33 | 50.63 | 55.37 | 86.37 | 70.03 | 68.61 | 71.31 | |
| ZERO+MaPLeBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 94.48 | 90.6 | 68.58 | 71.62 | 26.25 | 68.2 | 45.86 | 42.17 | 86.77 | 69.87 | 66.44 | — | |
| CoCoOpBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 94.43 | 90.14 | 65.32 | 71.88 | 22.94 | 67.36 | 45.73 | 39.23 | 83.97 | 68.44 | 64.94 | — | |
| CoCoOpTraining Source=ImageNet, Evaluation Mode=Cross-Dataset Transfer2026.03 | 94.43 | 90.14 | 65.32 | 71.88 | 22.94 | 67.36 | 45.73 | 45.37 | 86.06 | 68.21 | 65.74 | 71.02 | |
| CoCoOpYear=2022, Training Shots=16 shots, Source Training Dataset=ImageNet2026.03 | 94.43 | 90.14 | 65.32 | 71.88 | 22.94 | 67.36 | 45.73 | 45.37 | 86.06 | 68.21 | 65.74 | 71.02 | |
| CoCoOp2024.10 | 94.43 | 90.14 | 65.32 | 71.88 | 22.94 | 67.36 | 45.73 | 45.37 | 86.06 | 68.21 | 65.74 | 71.02 | |
| TDABackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 94.24 | 88.63 | 67.28 | 71.42 | 23.91 | 67.54 | 47.4 | 58 | 86.14 | 70.66 | 67.53 | — | |
| HPTYear=2024, Training Shots=16 shots, Source Training Dataset=ImageNet2026.03 | 94.2 | 92.63 | 66.33 | 74.84 | 25.68 | 68.75 | 50.87 | 47.36 | 86.21 | 70.5 | 67.74 | 71.72 | |
| TPTBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 94.16 | 87.79 | 66.87 | 68.98 | 24.78 | 65.5 | 47.75 | 42.44 | 84.67 | 68.04 | 65.1 | — | |
| PromptAlignBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 94.01 | 90.76 | 68.5 | 72.39 | 24.8 | 67.54 | 47.24 | 47.86 | 86.65 | 69.47 | 66.92 | — | |
| TCPTraining Source=ImageNet, Evaluation Mode=Cross-Dataset Transfer2026.03 | 93.97 | 91.25 | 64.69 | 71.21 | 23.45 | 67.15 | 44.35 | 51.45 | 86.69 | 68.73 | 66.29 | 71.4 | |
| TCP2024.10 | 93.97 | 91.25 | 64.69 | 71.21 | 23.45 | 67.15 | 44.35 | 51.45 | 86.69 | 68.73 | 66.29 | 71.4 | |
| ZERO+CoOpBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 93.85 | 88.36 | 64.9 | 67.23 | 19.14 | 64.73 | 43.62 | 33.53 | 82.67 | 66.61 | 62.46 | — | |
| MMATraining Source=ImageNet, Evaluation Mode=Cross-Dataset Transfer2026.03 | 93.8 | 90.3 | 66.13 | 72.07 | 25.33 | 68.17 | 46.57 | 49.24 | 86.12 | 68.32 | 66.61 | 71 | |
| MMAYear=2024, Training Shots=16 shots, Source Training Dataset=ImageNet2026.03 | 93.8 | 90.3 | 66.13 | 72.07 | 25.33 | 68.17 | 46.57 | 49.24 | 86.12 | 68.32 | 66.61 | 71 | |
| MMA2024.10 | 93.8 | 90.3 | 66.13 | 72.07 | 25.33 | 68.17 | 46.57 | 49.24 | 86.12 | 68.32 | 66.61 | 71 | |
| TPT+CoOpBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 93.75 | 88.93 | 67.06 | 68.25 | 25.89 | 66.4 | 47.15 | 48.78 | 83.82 | 66.53 | 65.66 | — | |
| CoOpBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 93.7 | 89.14 | 64.51 | 68.71 | 18.47 | 64.15 | 41.92 | 46.39 | 85.3 | 66.55 | 63.88 | — | |
| CoOpTraining Source=ImageNet, Evaluation Mode=Cross-Dataset Transfer2026.03 | 93.7 | 89.14 | 64.51 | 68.71 | 18.47 | 64.15 | 41.92 | 46.39 | 85.3 | 66.55 | 63.88 | 71.51 | |
| CoOpYear=2022, Training Shots=16 shots, Source Training Dataset=ImageNet2026.03 | 93.7 | 89.14 | 64.51 | 68.71 | 18.47 | 64.15 | 41.92 | 46.39 | 85.3 | 66.55 | 63.88 | 71.51 | |
| CLIP2024.10 | 93.7 | 89.1 | 65.7 | 70.7 | 24.9 | 62.6 | 44.3 | 48.3 | 85.9 | 67.6 | 65.24 | 66.7 | |
| CoOp2024.10 | 93.7 | 89.14 | 64.51 | 68.71 | 18.47 | 64.15 | 41.92 | 46.39 | 85.3 | 66.55 | 63.88 | 71.51 | |
| ZEROBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 93.66 | 87.75 | 68.04 | 67.68 | 25.21 | 65.03 | 46.12 | 34.33 | 86.53 | 67.77 | 64.21 | — | |
| C-TPTBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 93.6 | 88.2 | 65.8 | 69.8 | 24 | 64.8 | 46 | 43.2 | 83.7 | 65.7 | 64.48 | — | |
| PromptSRCTraining Source=ImageNet, Evaluation Mode=Cross-Dataset Transfer2026.03 | 93.6 | 90.25 | 65.7 | 70.25 | 23.9 | 67.1 | 46.87 | 45.5 | 86.15 | 68.75 | 65.81 | 71.27 | |
| P-SRCYear=2023, Training Shots=16 shots, Source Training Dataset=ImageNet2026.03 | 93.6 | 90.25 | 65.7 | 70.25 | 23.9 | 67.1 | 46.87 | 45.5 | 86.15 | 68.75 | 65.81 | 71.27 | |
| PromptSRC2024.10 | 93.6 | 90.25 | 65.7 | 70.25 | 23.9 | 67.1 | 46.87 | 45.5 | 86.15 | 68.75 | 65.81 | 71.27 | |
| TPT+MaPLeBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 93.59 | 90.72 | 66.5 | 72.37 | 24.7 | 67.54 | 45.87 | 47.8 | 86.64 | 69.19 | 66.5 | — | |
| MaPLeTraining Source=ImageNet, Evaluation Mode=Cross-Dataset Transfer2026.03 | 93.53 | 90.49 | 65.57 | 72.23 | 24.74 | 67.01 | 46.49 | 48.06 | 86.2 | 68.69 | 66.3 | 70.72 | |
| MaPLeYear=2023, Training Shots=16 shots, Source Training Dataset=ImageNet2026.03 | 93.53 | 90.49 | 65.57 | 72.23 | 24.74 | 67.01 | 46.49 | 48.06 | 86.2 | 68.69 | 66.3 | 70.72 | |
| CLIP-ViT-B/16Backbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 93.35 | 88.25 | 65.48 | 67.44 | 23.67 | 62.59 | 44.27 | 42.01 | 83.65 | 65.13 | 63.58 | — | |
| CLIPYear=2021, Training Shots=16 shots, Source Training Dataset=ImageNet2026.03 | 92.98 | 89.13 | 65.29 | 71.3 | 24.9 | 62.59 | 44.56 | 47.84 | 86.11 | 66.83 | 65.15 | 66.72 | |
| DiffTPTBackbone=ViT-B/16, Source Dataset=ImageNet, Prompt Tuning Shots=16-shot2025.01 | 92.49 | 88.22 | 67.01 | 70.1 | 25.6 | 65.74 | 47 | 43.13 | 87.23 | 62.67 | 64.92 | — |