Image Classification on 10 Fine-Grained Recognition Datasets (test)
58Aircraft AccuracyIsoCLIP
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| IsoCLIPIntra-modal=Yes, Classifier=NCM, Backbone=ViT-L/142026.03 | 58 | 97.2 | 87.8 | 72.9 | 88.5 | 99 | 90.8 | 94.4 | 75.8 | 86.9 | 85.1 | — | |
| Image-ImageIntra-modal=Yes, Classifier=NCM, Backbone=ViT-L/142026.03 | 55.5 | 96.8 | 83.7 | 70.4 | 84.7 | 98.8 | 90 | 91.8 | 76.7 | 85.9 | 83.4 | — | |
| IsoCLIPIntra-modal=Yes, Classifier=NCM, Backbone=ViT-B/162026.03 | 45.6 | 95.4 | 81.2 | 68.4 | 82.6 | 97.1 | 85.6 | 90.9 | 72 | 81.7 | 80.1 | — | |
| Image-ImageIntra-modal=Yes, Classifier=NCM, Backbone=ViT-B/162026.03 | 43.6 | 94.1 | 74.8 | 65.8 | 79.2 | 96.2 | 84 | 84.2 | 72.1 | 80.2 | 77.4 | — | |
| IsoCLIPIntra-modal=Yes, Classifier=NCM, Backbone=ViT-B/322026.03 | 35 | 93 | 73 | 66.8 | 80.6 | 93.8 | 78.6 | 84 | 70.5 | 77.8 | 75.3 | — | |
| Image-ImageIntra-modal=Yes, Classifier=NCM, Backbone=ViT-B/322026.03 | 34.5 | 91.9 | 67 | 64.4 | 77.7 | 93.7 | 77.3 | 77.1 | 70.2 | 76.6 | 73 | — | |
| ZEROBackbone=CLIP-ViT-B-16 + CLIP-ViT-L-142024.05 | 33.62 | 95.21 | 78.39 | 53.63 | 44.21 | 76.41 | 90.66 | 94.08 | 69.61 | 74.68 | 71.05 | 75.55 | |
| Image-TextIntra-modal=No, Classifier=Zero-Shot, Backbone=ViT-L/142026.03 | 32.5 | 94.8 | 76.8 | 53.1 | 58.1 | 79.3 | 91 | 93.6 | 67.6 | 74.2 | 72.1 | — | |
| Zero-ShotBackbone=CLIP-ViT-B-16 + CLIP-ViT-L-142024.05 | 30.54 | 94.04 | 76.16 | 51.83 | 54.38 | 75.76 | 88.03 | 92.86 | 66.96 | 73.7 | 70.43 | 74.73 | |
| DPEBackbone=ViT-B/162024.10 | 28.95 | 94.81 | 67.31 | 54.2 | 55.79 | 75.07 | 86.17 | 91.14 | 70.07 | 70.44 | 69.4 | — | |
| DPEBackbone=ViT-B/16, Category=Backpropagation-Based TTA, Venue=NIPS20242026.04 | 28.95 | 94.81 | 67.31 | 54.2 | 55.79 | 75.07 | 86.17 | 91.14 | 70.07 | 70.44 | 69.4 | — | |
| DPEBackbone=ViT-B/16, Optimization Source=Original paper2026.05 | 28.95 | 94.81 | 67.31 | 54.2 | 55.79 | 75.07 | 86.17 | 91.14 | 70.07 | 70.44 | 69.4 | — | |
| RLCFBackbone=CLIP-ViT-B-16 + CLIP-ViT-L-14, Tuned Parameters=theta_v, Adaptation Steps (t)=32024.05 | 28.54 | 94.73 | 70.93 | 53.27 | 47.41 | 71.74 | 87.28 | 91.15 | 69.38 | 73.24 | 68.77 | 71.34 | |
| BoostAdapterBackbone=ViT-B/16, Category=Backpropagation-Free TTA, Venue=NIPS20242026.04 | 27.45 | 94.77 | 69.3 | 45.69 | 61.22 | 71.66 | 87.17 | 89.51 | 68.09 | 71.93 | 68.68 | — | |
| HisTPTBackbone=ViT-B/16, Category=Backpropagation-Based TTA, Venue=NIPS20242026.04 | 26.9 | 94.5 | 69.2 | 48.9 | 49.7 | 71.2 | 89.3 | 89.1 | 67.2 | 70.1 | 67.6 | — | |
| CLIPTTABackbone=ViT-B/16, Optimization Source=Original paper2026.05 | 26.5 | 94.2 | 66.7 | 46.5 | 80.3 | 71.3 | 86.7 | 91.6 | 65.2 | 69.3 | 69.8 | — | |
| GS-BiasBackbone=ViT-B/16, Category=Backpropagation-Based TTA, Venue=ICML20252026.04 | 26.49 | 94.6 | 67.33 | 46.1 | 52.42 | 71.94 | 86.09 | 90.38 | 67.4 | 67.59 | 67.03 | — | |
| GRPO-TTABackbone=ViT-B/162026.05 | 26.43 | 95.56 | 74.05 | 51.89 | 58.89 | 75.84 | 88.12 | 90.82 | 69.35 | 73.59 | 70.45 | — | |
| ZEROBackbone=MaPLe2024.05 | 26.25 | 94.48 | 68.58 | 47.89 | 39.47 | 71.62 | 87.2 | 90.6 | 68.2 | 69.87 | 66.42 | 69.23 | |
| DOTABackbone=ViT-B/16, Optimization Source=Original paper2026.05 | 26.25 | 94.16 | 69.56 | 47.64 | 62.78 | 75.23 | 87.08 | 92.01 | 69.8 | 72.54 | 69.71 | — | |
| PTABackbone=ViT-B/16, Category=Backpropagation-Free TTA, Venue=Ours2026.04 | 26.1 | 94.81 | 68.55 | 47.7 | 61.57 | 75.23 | 86.44 | 91.06 | 69.21 | 73.12 | 69.38 | — | |
| DiffTPTBackbone=ViT-B/162024.10 | 25.6 | 92.49 | 67.01 | 47 | 43.13 | 70.1 | 87.23 | 88.22 | 65.74 | 62.67 | 65.47 | — | |
| DiffTPTBackbone=ViT-B/16, Category=Backpropagation-Based TTA, Venue=ICCV20232026.04 | 25.6 | 92.49 | 67.01 | 47 | 43.13 | 70.1 | 87.23 | 88.22 | 65.74 | 62.67 | 65.47 | — | |
| DiffTPTBackbone=ViT-B/16, Optimization Source=Original paper2026.05 | 25.6 | 92.49 | 67.01 | 47 | 43.13 | 70.1 | 87.23 | 88.22 | 65.74 | 62.67 | 64.92 | — | |
| ADAPTBackbone=ViT-B/16, Category=Backpropagation-Free TTA, Reproduced=true, Venue=NIPS20252026.04 | 25.59 | 93.39 | 68.56 | 46.57 | 63.06 | 73.28 | 85.72 | 91.58 | 69.09 | 70.58 | 68.74 | — | |
| RLCFBackbone=CLIP-ViT-B-16 + CLIP-ViT-L-14, Tuned Parameters=t_ctx, Adaptation Steps (t)=32024.05 | 25.4 | 95 | 72.91 | 51.93 | 45.96 | 72.49 | 86.84 | 89.55 | 69.04 | 72.31 | 68.14 | 72.4 | |
| RLCFBackbone=ViT-B/16, Optimization Source=Current study optimizations2026.05 | 25.35 | 93.84 | 67.29 | 48.04 | 45.15 | 70.28 | 85.12 | 89.45 | 66.34 | 69.28 | 66.01 | — | |
| TDABackbone=ViT-B/16, Category=Backpropagation-Free TTA, Reproduced=true, Venue=CVPR20242026.04 | 25.32 | 94.08 | 67.03 | 45.09 | 62.51 | 71.5 | 86.27 | 89.37 | 67.81 | 70.82 | 67.97 | — | |
| MTABackbone=ViT-B/16, Category=Backpropagation-Free TTA, Reproduced=true, Venue=CVPR20242026.04 | 25.23 | 94.32 | 66.43 | 44.68 | 47.75 | 71.34 | 86.25 | 89.37 | 66.82 | 67.33 | 65.95 | — | |
| ZEROBackbone=CLIP-ViT-B-162024.05 | 25.21 | 93.66 | 68.04 | 46.12 | 34.33 | 67.68 | 86.53 | 87.75 | 65.03 | 67.77 | 64.21 | 67.72 | |
| ZERO+EnsembleBackbone=CLIP-ViT-B-162024.05 | 25.21 | 94.41 | 68.97 | 45.86 | 42.17 | 67.17 | 86.77 | 87.83 | 67.63 | 69.18 | 65.52 | 68.3 | |
| SCABackbone=ViT-B/16, Category=Backpropagation-Free TTA, Reproduced=true, Venue=NIPS20252026.04 | 25.03 | 93.91 | 67.57 | 47.72 | 52.57 | 74.45 | 86.16 | 89.91 | 68.84 | 71.34 | 67.75 | — | |
| BCABackbone=ViT-B/16, Category=Backpropagation-Free TTA, Reproduced=true, Venue=CVPR20252026.04 | 24.87 | 93.91 | 66.8 | 44.62 | 49.23 | 70.77 | 86.04 | 89.02 | 66.57 | 67.12 | 65.9 | — | |
| TCABackbone=ViT-B/16, Category=Backpropagation-Free TTA, Venue=ICCV20252026.04 | 24.87 | 93.69 | 65.33 | 46.16 | 70.43 | 73.33 | 85.31 | 89.53 | 65.92 | 72.38 | 68.69 | — | |
| PromptAlignBackbone=MaPLe2024.05 | 24.8 | 94.01 | 68.5 | 47.24 | 47.86 | 72.39 | 86.65 | 90.76 | 67.54 | 69.47 | 66.92 | 68.99 | |
| Image-TextIntra-modal=No, Classifier=Zero-Shot, Backbone=ViT-B/162026.03 | 24.8 | 92.9 | 65.3 | 44.4 | 47.8 | 71.4 | 86.1 | 89.1 | 62.6 | 66.7 | 65.1 | — | |
| TPTBackbone=ViT-B/162024.10 | 24.78 | 94.16 | 66.87 | 47.75 | 42.44 | 68.98 | 84.67 | 87.79 | 65.5 | 68.04 | 65.1 | — | |
| CLIP-ViT-B/16Backbone=ViT-B/16, Category=Zero-Shot, Reproduced=true, Venue=ICML20212026.04 | 24.78 | 94.12 | 65.91 | 44.39 | 47.69 | 71.34 | 86.08 | 89.1 | 66.31 | 66.72 | 65.64 | — | |
| TPTBackbone=ViT-B/16, Category=Backpropagation-Based TTA, Venue=NIPS20222026.04 | 24.78 | 94.16 | 68.5 | 47.75 | 42.44 | 68.98 | 84.67 | 87.79 | 65.5 | 68.04 | 65.1 | — | |
| TPTBackbone=ViT-B/16, Optimization Source=Original paper2026.05 | 24.78 | 94.16 | 66.87 | 47.75 | 42.44 | 68.98 | 84.67 | 87.79 | 65.5 | 68.04 | 65.1 | — | |
| Zero-ShotBackbone=MaPLe2024.05 | 24.74 | 93.53 | 65.57 | 46.49 | 48.06 | 72.23 | 86.2 | 90.49 | 67.01 | 68.69 | 66.3 | 67.85 | |
| TPTBackbone=MaPLe2024.05 | 24.7 | 93.59 | 66.5 | 45.87 | 47.8 | 72.37 | 86.64 | 90.72 | 67.54 | 69.19 | 66.49 | 68.36 | |
| WATTBackbone=ViT-B/16, Optimization Source=Original paper2026.05 | 24.24 | 93.59 | 66.78 | 46.87 | 51.95 | 68.53 | 84.81 | 88.14 | 65.84 | 65.74 | 65.65 | — | |
| RLCFBackbone=CLIP-ViT-B-16 + CLIP-ViT-L-14, Tuned Parameters=theta_v, Adaptation Steps (t)=12024.05 | 24.06 | 94.65 | 63.12 | 52.21 | 47.74 | 72.56 | 86.9 | 89.51 | 68.5 | 71.49 | 67.07 | 70 | |
| TDABackbone=ViT-B/162024.10 | 23.91 | 94.24 | 67.28 | 47.4 | 58 | 71.42 | 86.14 | 88.63 | 67.62 | 70.66 | 67.53 | — | |
| TDABackbone=ViT-B/16, Optimization Source=Original paper2026.05 | 23.91 | 94.24 | 67.28 | 47.4 | 58 | 71.42 | 86.14 | 88.63 | 67.62 | 70.66 | 67.53 | — | |
| RLCFBackbone=CLIP-ViT-B-16 + CLIP-ViT-L-14, Tuned Parameters=t_ctx, Adaptation Steps (t)=12024.05 | 23.71 | 94.09 | 69.76 | 50.34 | 46.87 | 71.58 | 85.9 | 89.01 | 67.33 | 69.84 | 66.84 | 69.8 | |
| CLIPBackbone=ViT-B/162024.10 | 23.67 | 93.35 | 65.48 | 44.27 | 42.01 | 67.44 | 83.65 | 88.25 | 62.59 | 65.13 | 63.58 | — | |
| Zero-ShotBackbone=CLIP-ViT-B-162024.05 | 23.67 | 93.35 | 65.48 | 44.27 | 42.01 | 67.44 | 83.65 | 88.25 | 62.59 | 65.13 | 63.58 | 65.31 | |
| EnsembleBackbone=ViT-B/162024.10 | 23.22 | 93.55 | 66.11 | 45.04 | 50.42 | 66.99 | 82.86 | 86.92 | 65.63 | 65.16 | 64.59 | — | |
| EnsembleBackbone=CLIP-ViT-B-162024.05 | 23.22 | 93.91 | 66.16 | 45.09 | 50.42 | 67.07 | 84.04 | 88.28 | 66.26 | 67.51 | 65.2 | 66.66 | |
| CLIPBackbone=ViT-B/16, Optimization Source=Original paper2026.05 | 23.22 | 93.55 | 66.11 | 45.04 | 50.42 | 66.99 | 82.86 | 86.92 | 65.63 | 65.16 | 64.59 | — | |
| TPTBackbone=CLIP-ViT-B-162024.05 | 23.13 | 93.93 | 66.68 | 47.04 | 42.86 | 68.75 | 84.67 | 87.23 | 65.39 | 68.16 | 64.78 | 67.42 | |
| CoCoOpBackbone=ViT-B/16, Optimization Source=Original paper2026.05 | 22.29 | 93.79 | 64.9 | 45.45 | 39.23 | 70.85 | 83.97 | 90.46 | 66.89 | 68.44 | 64.63 | — | |
| DPEBackbone=ResNet-502024.10 | 19.8 | 90.83 | 59.26 | 50.18 | 41.67 | 67.6 | 77.83 | 85.97 | 64.23 | 61.98 | 61.93 | — | |
| Image-TextIntra-modal=No, Classifier=Zero-Shot, Backbone=ViT-B/322026.03 | 19.1 | 91.2 | 60.4 | 43.6 | 45.2 | 67 | 80.5 | 87.5 | 62 | 62 | 61.9 | — | |
| CoOpBackbone=ViT-B/162024.10 | 18.47 | 93.7 | 64.51 | 41.92 | 46.39 | 68.71 | 85.3 | 89.14 | 64.15 | 66.55 | 63.88 | — | |
| CoOpBackbone=ViT-B/16, Optimization Source=Original paper2026.05 | 18.47 | 93.7 | 64.51 | 41.92 | 46.39 | 68.71 | 85.3 | 89.14 | 64.15 | 66.55 | 63.88 | — | |
| TDABackbone=ResNet-502024.10 | 17.61 | 89.7 | 57.78 | 43.74 | 42.11 | 68.74 | 77.75 | 86.18 | 62.53 | 64.18 | 61.03 | — | |
| DiffTPTBackbone=ResNet-502024.10 | 17.6 | 86.89 | 60.71 | 40.72 | 41.04 | 63.53 | 79.21 | 83.4 | 62.72 | 62.67 | 59.85 | — | |
| TPTBackbone=ResNet-502024.10 | 17.58 | 87.02 | 58.46 | 40.84 | 28.33 | 62.69 | 74.88 | 84.49 | 61.46 | 60.82 | 57.66 | — | |
| EnsembleBackbone=ResNet-502024.10 | 16.11 | 87.26 | 55.89 | 40.37 | 25.79 | 62.77 | 74.82 | 82.97 | 60.85 | 59.48 | 56.63 | — | |
| CLIPBackbone=ResNet-502024.10 | 15.66 | 85.88 | 55.7 | 40.37 | 23.69 | 61.75 | 73.97 | 83.57 | 58.8 | 58.84 | 55.82 | — | |
| CoOpBackbone=ResNet-502024.10 | 15.12 | 86.53 | 55.32 | 37.29 | 26.2 | 61.55 | 75.59 | 87 | 58.15 | 59.05 | 56.18 | — |