OOD Detection on ImageNet 1k
9.81Average FPR95TANL
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| TANLTraining or Adaptation Strategy=Training-free & test-time adaptation, Backbone=ViT-B/162026.03 | 9.81 | 0.42 | 99.84 | 3.53 | 99.07 | 21.9 | 95.87 | 13.38 | 97.11 | 97.97 | |
| SynOODTraining or Adaptation Strategy=training (or fine-tuning), Backbone=ViT-B/162026.03 | 14.27 | 1.57 | 99.57 | 20.46 | 95.82 | 12.12 | 93.37 | 22.94 | 95.29 | 97.01 | |
| AdaNDTraining or Adaptation Strategy=training (or fine-tuning), Backbone=ViT-B/162026.03 | 16 | 1.91 | 99.49 | 20.53 | 95.49 | 20.95 | 94.55 | 21.76 | 93.01 | 95.58 | |
| CSPTraining or Adaptation Strategy=Training-free & non-adaptive, Backbone=ViT-B/162026.03 | 17.51 | 1.54 | 99.6 | 13.66 | 96.66 | 29.32 | 92.9 | 25.52 | 93.86 | 95.76 | |
| OODDTraining or Adaptation Strategy=Training-free & test-time adaptation, Backbone=ViT-B/162026.03 | 18.79 | 0.85 | 99.79 | 12.94 | 97.17 | 30.68 | 92.51 | 30.67 | 94.51 | 96 | |
| AdaNegTraining or Adaptation Strategy=Training-free & test-time adaptation, Backbone=ViT-B/162026.03 | 18.92 | 0.59 | 99.71 | 9.5 | 97.44 | 34.34 | 94.55 | 31.27 | 94.93 | 96.66 | |
| CMATraining or Adaptation Strategy=training (or fine-tuning), Backbone=ViT-B/162026.03 | 19.93 | 1.65 | 99.62 | 16.84 | 96.36 | 27.65 | 93.11 | 33.58 | 91.64 | 95.13 | |
| DualCnstTraining or Adaptation Strategy=Training-free & non-adaptive, Backbone=ViT-B/162026.03 | 21.45 | 0.98 | 99.7 | 18.13 | 95.66 | 31.77 | 91.81 | 34.91 | 91.73 | 94.72 | |
| KR-NFTTraining or Adaptation Strategy=training (or fine-tuning), Backbone=ViT-B/162026.03 | 22.79 | 0.82 | 99.62 | 17.83 | 96.15 | 36.12 | 92.64 | 36.38 | 91.96 | 95.09 | |
| NegPromptTraining or Adaptation Strategy=training (or fine-tuning), Backbone=ViT-B/162026.03 | 23.01 | 6.32 | 98.73 | 22.89 | 95.55 | 27.6 | 93.34 | 35.21 | 91.6 | 94.81 | |
| LAPTTraining or Adaptation Strategy=training (or fine-tuning), Backbone=ViT-B/162026.03 | 23.4 | 1.16 | 99.63 | 19.12 | 96.01 | 33.01 | 92.01 | 40.32 | 91.06 | 94.68 | |
| NegLabelTraining or Adaptation Strategy=Training-free & non-adaptive, Backbone=ViT-B/162026.03 | 25.4 | 1.91 | 99.49 | 20.53 | 95.49 | 35.59 | 91.64 | 43.56 | 90.22 | 94.21 | |
| LoCoOpTraining or Adaptation Strategy=training (or fine-tuning), Backbone=ViT-B/162026.03 | 28.66 | 16.05 | 96.86 | 23.44 | 95.07 | 32.87 | 91.98 | 42.28 | 90.19 | 93.52 | |
| CATEXBackbone=CLIP-B/16, Mode=Prompt Tuning, Context Tokens=162024.07 | 29.66 | 10.18 | 97.88 | 33.87 | 92.83 | 41.43 | 90.48 | 33.17 | 92.73 | 93.48 | |
| EOETraining or Adaptation Strategy=Training-free & non-adaptive, Backbone=ViT-B/162026.03 | 30.09 | 12.29 | 97.52 | 20.04 | 95.73 | 30.16 | 92.95 | 57.53 | 85.64 | 92.96 | |
| CLIPNTraining or Adaptation Strategy=training (or fine-tuning), Backbone=ViT-B/162026.03 | 31.1 | 23.94 | 95.27 | 26.17 | 93.93 | 33.45 | 92.28 | 40.83 | 90.93 | 93.1 | |
| CMA (Training-free)Training or Adaptation Strategy=Training-free & non-adaptive, Backbone=ViT-B/162026.03 | 32.79 | 23.84 | 96.89 | 30.11 | 93.69 | 29.86 | 93.17 | 47.35 | 88.47 | 93.05 | |
| CoVerTraining or Adaptation Strategy=Training-free & non-adaptive, Backbone=ViT-B/162026.03 | 34.88 | 22.55 | 95.98 | 32.85 | 93.42 | 40.71 | 90.27 | 43.39 | 90.14 | 92.45 | |
| NPOSMode=Fine-tuning2024.07 | 37.93 | 16.58 | 96.19 | 43.77 | 90.44 | 45.27 | 89.44 | 46.12 | 88.8 | 91.22 | |
| KNNMode=Fine-tuning2024.07 | 42.19 | 29.17 | 94.52 | 35.62 | 92.67 | 39.61 | 91.02 | 64.35 | 85.67 | 90.97 | |
| VOSMode=Fine-tuning2024.07 | 43.24 | 31.65 | 94.53 | 43.03 | 91.92 | 41.62 | 90.23 | 56.67 | 86.74 | 90.86 | |
| MCMBackbone=CLIP-B/16, Mode=Zero-shot2024.07 | 43.55 | 32.08 | 94.41 | 39.21 | 92.28 | 44.88 | 89.83 | 58.05 | 85.96 | 90.62 | |
| MCMTraining or Adaptation Strategy=Training-free & non-adaptive, Backbone=ViT-B/162026.03 | 43.93 | 32.2 | 94.59 | 38.8 | 92.25 | 46.2 | 90.31 | 58.5 | 86.12 | 90.82 | |
| EnergyMode=Fine-tuning2024.07 | 47.67 | 29.75 | 94.68 | 53.18 | 87.33 | 56.4 | 85.6 | 51.35 | 88 | 88.9 | |
| ViMMode=Fine-tuning2024.07 | 50.2 | 32.19 | 93.16 | 54.01 | 87.19 | 60.67 | 83.75 | 53.94 | 87.18 | 87.82 | |
| SCALEArchitecture=ResNet2026.03 | 59.76 | — | — | — | — | — | — | — | — | 81.36 | |
| DOEMode=Fine-tuning2024.07 | 59.83 | 55.87 | 85.98 | 80.94 | 76.26 | 67.84 | 83.05 | 34.67 | 88.9 | 83.54 | |
| ASHArchitecture=ResNet2026.03 | 63.32 | — | — | — | — | — | — | — | — | 78.17 | |
| SSD+Mode=Fine-tuning2024.07 | 64.55 | 59.6 | 85.54 | 75.62 | 73.8 | 83.6 | 68.11 | 39.4 | 82.4 | 77.46 | |
| RMDSArchitecture=ResNet2026.03 | 65.04 | — | — | — | — | — | — | — | — | 76.99 | |
| GENArchitecture=ResNet2026.03 | 65.32 | — | — | — | — | — | — | — | — | 76.85 | |
| MSPArchitecture=ResNet2026.03 | 65.68 | — | — | — | — | — | — | — | — | 76.02 | |
| AggBBAS-ECArchitecture=ResNet2026.03 | 66.17 | — | — | — | — | — | — | — | — | 78.6 | |
| MSPMode=Fine-tuning2024.07 | 67.31 | 54.05 | 87.43 | 73.37 | 78.03 | 72.98 | 78.03 | 68.85 | 79.06 | 80.64 | |
| AggBBAS-EDArchitecture=ResNet2026.03 | 68.03 | — | — | — | — | — | — | — | — | 77.9 | |
| KNNArchitecture=ResNet2026.03 | 70.87 | — | — | — | — | — | — | — | — | 71.1 | |
| VIMArchitecture=ResNet2026.03 | 71.35 | — | — | — | — | — | — | — | — | 72.08 | |
| BBAS-ECArchitecture=ResNet2026.03 | 71.9 | — | — | — | — | — | — | — | — | 75.54 | |
| ODINArchitecture=ResNet2026.03 | 72.5 | — | — | — | — | — | — | — | — | 74.75 | |
| SHEArchitecture=ResNet2026.03 | 73.01 | — | — | — | — | — | — | — | — | 73.78 | |
| BBAS-EDArchitecture=ResNet2026.03 | 73.77 | — | — | — | — | — | — | — | — | 74.31 | |
| ZOCTraining or Adaptation Strategy=training (or fine-tuning), Backbone=ViT-B/162026.03 | 85.19 | 87.3 | 86.09 | 81.51 | 81.2 | 73.06 | 83.39 | 98.9 | 76.46 | 81.79 | |
| MDSArchitecture=ResNet2026.03 | 85.45 | — | — | — | — | — | — | — | — | 55.44 | |
| MDSEnsArchitecture=ResNet2026.03 | 93.52 | — | — | — | — | — | — | — | — | 49.67 |