Fine-grained classification on Aircraft (test)
95.4AccuracySR-GNN
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SR-GNN2022.09 | 95.4 | — | — | |
| Proposed (CAP)Transfer/Joint Learning Strategy=false2021.01 | 94.9 | — | — | |
| SotA2022.09 | 94.9 | — | — | |
| DCLTransfer/Joint Learning Strategy=false2021.01 | 93 | — | — | |
| MCLossTransfer/Joint Learning Strategy=false2021.01 | 92.9 | — | — | |
| GPipeTransfer/Joint Learning Strategy=false2021.01 | 92.7 | — | — | |
| BARMTransfer/Joint Learning Strategy=false2021.01 | 92.5 | — | — | |
| DFLTransfer/Joint Learning Strategy=false2021.01 | 92 | — | — | |
| SaSPA w/o BLIP-diffusionType=Ours, Backbone=ResNet-50, Diffusion Model=Stable Diffusion v1.52024.06 | 87.4 | — | — | |
| SaSPAType=Ours, Backbone=ResNet-50, Diffusion Model=BLIP-diffusion2024.06 | 86.6 | — | — | |
| CAL-AugType=Traditional, Backbone=ResNet-502024.06 | 84.9 | — | — | |
| Real GuidanceType=Generative, Backbone=ResNet-502024.06 | 84.8 | — | — | |
| CAL-Aug + CutMixType=Traditional, Backbone=ResNet-502024.06 | 84.5 | — | — | |
| RandAug + CutMixType=Traditional, Backbone=ResNet-502024.06 | 84 | — | — | |
| RandAugType=Traditional, Backbone=ResNet-502024.06 | 83.7 | — | — | |
| ALIAType=Generative, Backbone=ResNet-502024.06 | 83.1 | — | — | |
| CutMixType=Traditional, Backbone=ResNet-502024.06 | 81.8 | — | — | |
| No AugType=Traditional, Backbone=ResNet-502024.06 | 81.4 | — | — | |
| MTACategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 20.3 | — | 2.4 | |
| R-TPTCategory=Tuning-based, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 19.1 | — | 9.5 | |
| CLIPCategory=Baseline, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 18.2 | — | 0 | |
| MACCategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 18.2 | — | 15.5 | |
| TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 16.9 | — | 0 | |
| SS-TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 16.6 | — | 13.3 | |
| EnsembleBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 16.4 | — | 6.3 | |
| TAPTCategory=Tuning-based, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 16.3 | — | 2.6 | |
| CLIPBackbone=CLIP-ResNet50, Epsilon=1.02026.06 | 15.7 | — | 0 | |
| R-TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 15.7 | — | 11 | |
| TTCCategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 15.3 | — | 0.7 | |
| DOCBackbone=CLIP-ResNet50, Epsilon=1.0, Effective batch size=12026.06 | 12.5 | — | 0.8 | |
| TTCBackbone=CLIP-ResNet50, Epsilon=1.0, Effective batch size=12026.06 | 12.1 | — | 0.7 |