Fine-Grained Visual Classification on FGVC Aircraft
94.1Top-1 AccuracyCCFR
Evaluation results
| Method | Links | |
|---|---|---|
| CCFRBase Model=ResNet-50, re-ranking=true2021.02 | 94.1 | |
| DF-GMMBase Model=ResNet-502021.02 | 93.8 | |
| PMGBase Model=ResNet-502021.02 | 93.4 | |
| API-NetBackbone=ResNet-101, Input size=448 x 4482021.01 | 93.4 | |
| GCLBase Model=ResNet-502021.02 | 93.2 | |
| DATL + WS-DAN2020.10 | 93.1 | |
| WS-DAN2020.10 | 93 | |
| DCLBase Model=ResNet-502021.02 | 93 | |
| API-netBase Model=ResNet-502021.02 | 93 | |
| CCFRBase Model=ResNet-50, re-ranking=false2021.02 | 93 | |
| API-NetBackbone=ResNet-50, Input size=448 x 4482021.01 | 93 | |
| EfficientNet B7Model=EfficientNet B72020.01 | 92.9 | |
| CINBackbone=ResNet-101, Input size=448 x 4482021.01 | 92.8 | |
| PCA-NetBackbone=ResNet-101, Input size=448 x 4482021.01 | 92.8 | |
| DLA2020.10 | 92.6 | |
| MC-LossBackbone=ResNet-50, Input size=448 x 4482021.01 | 92.6 | |
| Cross-XBackbone=ResNet-50, Input size=448 x 4482021.01 | 92.6 | |
| CINBackbone=ResNet-50, Input size=448 x 4482021.01 | 92.6 | |
| Assemble-ResNet-FGVC-50Model=Assemble-ResNet-FGVC-502020.01 | 92.4 | |
| ACNetBase Model=ResNet-502021.02 | 92.4 | |
| ACNetBackbone=ResNet-50, Input size=448 x 4482021.01 | 92.4 | |
| PCA-NetBackbone=ResNet-50, Input size=448 x 4482021.01 | 92.4 | |
| DFL-CNN2020.10 | 92 | |
| Mix+Base Model=ResNet-502021.02 | 92 | |
| DFL-CNNBackbone=ResNet-50, Input size=448 x 4482021.01 | 91.7 | |
| MPN-COV2020.10 | 91.4 | |
| NTSBase Model=ResNet-502021.02 | 91.4 | |
| NTS-NetBackbone=ResNet-50, Input size=448 x 4482021.01 | 91.4 | |
| Subset B2020.10 | 90.7 | |
| Inception-v3Base Model=Inception-v32021.02 | 90.7 | |
| BCNBase Model=ResNet-502021.02 | 90.3 | |
| MA-CNN2020.10 | 89.9 | |
| MA-CNNBase Model=VGG-192021.02 | 89.9 | |
| MA-CNNBackbone=VGG-19, Input size=448 x 4482021.01 | 89.9 | |
| GP-2562020.10 | 89.8 | |
| Improved Bilinear-CNN2020.10 | 88.5 | |
| FT ResNet-50Backbone=ResNet-50, Input size=448 x 4482021.01 | 88.5 | |
| Spatial-RNNBase Model=M-Net/D-Net2021.02 | 88.4 | |
| FT VGG-19Backbone=VGG-19, Input size=448 x 4482021.01 | 84.8 | |
| Bilinear-CNN2020.10 | 84.1 | |
| DINOv3-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 81.97 | |
| ResNet-50Model=ResNet-502020.01 | 78.8 | |
| VECA-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 256, Core budget (C)=642026.05 | 74.47 | |
| DINOv2-BEvaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 74.02 | |
| DINOv2-Reg-BEvaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 70.78 | |
| VECA-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 256, Core budget (C)=82026.05 | 70.72 | |
| DFN2B-CLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 66.49 | |
| OpenCLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 62.89 | |
| SigLIP2-B/16-256Evaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 62.8 | |
| CLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 56.65 | |
| C-RADIOv3-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 52.18 |