Image Classification on CARS196 (test)
91.5AccuracyNED
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| NED2020.06 | 91.5 | 1.5 | — | — | — | |
| WkNNvariant=[9]2020.06 | 91.3 | 12.2 | — | — | — | |
| WkNNvariant=[16]2020.06 | 91.3 | 12.3 | — | — | — | |
| CGC2021.10 | 90.28 | — | — | — | — | |
| KNN2020.06 | 90.1 | 10 | — | — | — | |
| Baseline2021.10 | 89.71 | — | — | — | — | |
| 1NNsource=reported2020.06 | 89.1 | — | — | — | — | |
| 1NNsource=re-implemented2020.06 | 89.1 | — | — | — | — | |
| NEDAttack Type=PGD, epsilon=0.12020.06 | 89 | 2 | — | — | — | |
| WkNNcitation=[9], Attack Type=PGD, epsilon=0.12020.06 | 88.9 | 12.8 | — | — | — | |
| WkNNcitation=[16], Attack Type=PGD, epsilon=0.12020.06 | 88.9 | 12.9 | — | — | — | |
| NEDAttack Type=FGSM, epsilon=0.12020.06 | 88.7 | 1.9 | — | — | — | |
| WkNNcitation=[9], Attack Type=FGSM, epsilon=0.12020.06 | 88.6 | 12.7 | — | — | — | |
| WkNNcitation=[16], Attack Type=FGSM, epsilon=0.12020.06 | 88.6 | 12.9 | — | — | — | |
| WkNNcitation=[9], Attack Type=BIM, epsilon=0.12020.06 | 88.4 | 13.2 | — | — | — | |
| WkNNcitation=[16], Attack Type=BIM, epsilon=0.12020.06 | 88.4 | 13.3 | — | — | — | |
| NEDAttack Type=BIM, epsilon=0.12020.06 | 88.4 | 2.2 | — | — | — | |
| kNNAttack Type=PGD, epsilon=0.12020.06 | 87.8 | 10.6 | — | — | — | |
| kNNAttack Type=FGSM, epsilon=0.12020.06 | 87.5 | 10.4 | — | — | — | |
| kNNAttack Type=BIM, epsilon=0.12020.06 | 87.3 | 10.8 | — | — | — | |
| 1NNAttack Type=PGD, epsilon=0.12020.06 | 86.4 | — | — | — | — | |
| 1NNAttack Type=FGSM, epsilon=0.12020.06 | 86.3 | — | — | — | — | |
| 1NNAttack Type=BIM, epsilon=0.12020.06 | 85.7 | — | — | — | — | |
| NEDAttack Type=FGSM, epsilon=0.32020.06 | 81.3 | 3.6 | — | — | — | |
| WkNNcitation=[9], Attack Type=FGSM, epsilon=0.32020.06 | 80.9 | 12.7 | — | — | — | |
| WkNNcitation=[16], Attack Type=FGSM, epsilon=0.32020.06 | 80.9 | 12.9 | — | — | — | |
| kNNAttack Type=FGSM, epsilon=0.32020.06 | 79.6 | 9.8 | — | — | — | |
| NEDAttack Type=PGD, epsilon=0.32020.06 | 78 | 2.9 | — | — | — | |
| 1NNAttack Type=FGSM, epsilon=0.32020.06 | 77.7 | — | — | — | — | |
| WkNNcitation=[9], Attack Type=PGD, epsilon=0.32020.06 | 77.7 | 14.5 | — | — | — | |
| WkNNcitation=[16], Attack Type=PGD, epsilon=0.32020.06 | 77.7 | 14.5 | — | — | — | |
| kNNAttack Type=PGD, epsilon=0.32020.06 | 76.1 | 10.9 | — | — | — | |
| 1NNAttack Type=PGD, epsilon=0.32020.06 | 74.4 | — | — | — | — | |
| NEDAttack Type=BIM, epsilon=0.32020.06 | 73.3 | 2.5 | — | — | — | |
| WkNNcitation=[16], Attack Type=BIM, epsilon=0.32020.06 | 73.2 | 13.2 | — | — | — | |
| WkNNcitation=[9], Attack Type=BIM, epsilon=0.32020.06 | 73 | 13.1 | — | — | — | |
| kNNAttack Type=BIM, epsilon=0.32020.06 | 71.8 | 9.6 | — | — | — | |
| 1NNAttack Type=BIM, epsilon=0.32020.06 | 69.5 | — | — | — | — | |
| C x SEx (Using training-set Examples)=true, Img (Comparing at Image level)=true, Patch (Comparing at Patch level)=false, R (Re-ranking)=true, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 91.06 | — | — | |
| CHM-CorrEx (Using training-set Examples)=true, Img (Comparing at Image level)=true, Patch (Comparing at Patch level)=true, R (Re-ranking)=false, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 85.03 | — | — | |
| CLIP ViT-L/14Evaluation Protocol=Linear Probe, Backbone=ViT-L/142025.12 | — | — | 59.82 | 79.41 | 85.38 | |
| CLIP ViT-L/14Evaluation Protocol=Few-shot, Backbone=ViT-L/142025.12 | — | — | 55.41 | 81.18 | 86.04 | |
| DIOR (Ours)Evaluation Protocol=Linear Probe, Prompting Strategy=conditional, Base LLM=Llama-3.22025.12 | — | — | 80.02 | 89.67 | 93.15 | |
| DIOR (Ours)Evaluation Protocol=Few-shot, Prompting Strategy=conditional, Base LLM=Llama-3.22025.12 | — | — | 77.47 | 91.33 | 93.38 | |
| DIOR (w/o condition)Evaluation Protocol=Linear Probe, Prompting Strategy=non-conditional2025.12 | — | — | 40.06 | 58.83 | 69.53 | |
| DIOR (w/o condition)Evaluation Protocol=Few-shot, Prompting Strategy=non-conditional2025.12 | — | — | 39.44 | 64.14 | 72.43 | |
| EMD-CorrEx (Using training-set Examples)=true, Img (Comparing at Image level)=true, Patch (Comparing at Patch level)=true, R (Re-ranking)=false, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 87.4 | — | — | |
| k-NN + cosineEx (Using training-set Examples)=true, Img (Comparing at Image level)=true, Patch (Comparing at Patch level)=false, R (Re-ranking)=false, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 87.48 | — | — | |
| k-NN + SEx (Using training-set Examples)=true, Img (Comparing at Image level)=true, Patch (Comparing at Patch level)=false, R (Re-ranking)=false, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 88.9 | — | — | |
| PIPNetEx (Using training-set Examples)=false, Img (Comparing at Image level)=false, Patch (Comparing at Patch level)=true, R (Re-ranking)=false, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 86.5 | — | — | |
| ProtoKNNEx (Using training-set Examples)=true, Img (Comparing at Image level)=false, Patch (Comparing at Patch level)=true, R (Re-ranking)=false, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 90.2 | — | — | |
| ProtoPNetEx (Using training-set Examples)=false, Img (Comparing at Image level)=false, Patch (Comparing at Patch level)=true, R (Re-ranking)=false, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 85.31 | — | — | |
| ProtoPoolEx (Using training-set Examples)=false, Img (Comparing at Image level)=false, Patch (Comparing at Patch level)=true, R (Re-ranking)=false, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 88.9 | — | — | |
| ProtoPShareEx (Using training-set Examples)=false, Img (Comparing at Image level)=false, Patch (Comparing at Patch level)=true, R (Re-ranking)=false, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 86.4 | — | — | |
| ProtoTreeEx (Using training-set Examples)=false, Img (Comparing at Image level)=false, Patch (Comparing at Patch level)=true, R (Re-ranking)=false, Backbone=ResNet-50, Pre-training=ImageNet2023.08 | — | — | 86.6 | — | — |