Fine-grained Image Classification on CUB-200 2011
92.6AccuracyCSQA-Net
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CSQA-NetPublication=n/a, Backbone=Swin-B2024.03 | 92.6 | — | — | — | |
| DCALPublication=CVPR 2022, Backbone=R50-ViT-B2024.03 | 92 | — | — | — | |
| Dual-TRPublication=TCSVT 2023, Backbone=ViT-B-162024.03 | 92 | — | — | — | |
| MetaFormer-1Backbone=MetaFormer-1, Input in Testing=image+text, Training Input Modality=Image+Text2022.03 | 91.9 | — | — | — | |
| CAPBackbone=Xception, Input in Testing=image2022.03 | 91.8 | — | — | — | |
| MetaFormer-1Backbone=MetaFormer-1, Input in Testing=image+attr, Training Input Modality=Image+Attribute2022.03 | 91.8 | — | — | — | |
| IELTPublication=TMM 2023, Backbone=ViT-B-162024.03 | 91.8 | — | — | — | |
| ACC-ViTPublication=PR 2024, Backbone=ViT-B-162024.03 | 91.8 | — | — | — | |
| TransFGBackbone=ViT-B_162021.03 | 91.7 | — | — | — | |
| MetaFormer-1Backbone=MetaFormer-1, Input in Testing=image, Training Input Modality=Image+Text2022.03 | 91.7 | — | — | — | |
| TransFGPublication=AAAI 2022, Backbone=ViT-B-162024.03 | 91.7 | — | — | — | |
| ViT-NeTPublication=ICML 2022, Backbone=Swin-B2024.03 | 91.6 | — | — | — | |
| MetaFormer-1Backbone=MetaFormer-1, Input in Testing=image, Training Input Modality=Image+Attribute2022.03 | 91.5 | — | — | — | |
| MetaFormer-1Backbone=MetaFormer-1, Input in Testing=image, Training Input Modality=Image-Only2022.03 | 91.4 | — | — | — | |
| AA-transPublication=PR 2023, Backbone=ViT-B-162024.03 | 91.4 | — | — | — | |
| RAMS-TransBackbone=ViT-B_162021.07 | 91.3 | — | — | — | |
| Ours_PMGLabel Setting=Multi-granularity2020.11 | 91.25 | 97.98 | 93.5 | 82.26 | |
| DATL + WS-DAN2020.10 | 91.2 | — | — | — | |
| CCFRBase Model=ResNet-50, re-ranking=true2021.02 | 91.1 | — | — | — | |
| TransIFC+Publication=TMM 2023, Backbone=Swin-B2024.03 | 91 | — | — | — | |
| TransFG & PSMBackbone=ViT-B_162021.07 | 90.9 | — | — | — | |
| CCFRBase Model=ResNet-50, re-ranking=false2021.02 | 90.7 | — | — | — | |
| ViTBackbone=ViT-B_162021.07 | 90.6 | — | — | — | |
| TA-CFNPublication=TNNLS 2023, Backbone=ResNet-502024.03 | 90.5 | — | — | — | |
| CSQA-NetPublication=n/a, Backbone=ResNet-502024.03 | 90.5 | — | — | — | |
| Stacked LSTMBase Model=ResNet502020.03 | 90.4 | — | — | — | |
| StackedLSTMBase Model=GoogleNet2021.02 | 90.4 | — | — | — | |
| StackedLSTMBackbone=GoogleNet2021.07 | 90.4 | — | — | — | |
| WPS-CPMBase Model=GoogLeNet + ResNet50, Model Component=GoogLeNet + D+A2020.02 | 90.4 | — | — | — | |
| StackedLSTMBackbone=GoogleNet2021.03 | 90.4 | — | — | — | |
| StackedLSTMBackbone=GoogleNet, Input in Testing=image2022.03 | 90.4 | — | — | — | |
| ViTBackbone=ViT-B_162021.03 | 90.3 | — | — | — | |
| Mix+Base Model=ResNet-502021.02 | 90.2 | — | — | — | |
| P2P-NetPublication=CVPR 2022, Backbone=ResNet-502024.03 | 90.2 | — | — | — | |
| API-NetBackbone=DenseNet-1612021.07 | 90 | — | — | — | |
| API-NetBackbone=DenseNet-1612021.03 | 90 | — | — | — | |
| DeiTBackbone=DeiT-B2021.03 | 90 | — | — | — | |
| API-NetPublication=AAAI 2020, Backbone=DenseNet-1612024.03 | 90 | — | — | — | |
| GDSMP-NetPublication=PR 2023, Backbone=ResNet-502024.03 | 89.9 | — | — | — | |
| DINOv3-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 89.82 | — | — | — | |
| PMGBase Model=ResNet50, Evaluation Mode=Combined Accuracy2020.03 | 89.6 | — | — | — | |
| Subset B2020.10 | 89.6 | — | — | — | |
| Inception-v3Base Model=Inception-v32021.02 | 89.6 | — | — | — | |
| PMGBase Model=ResNet-502021.02 | 89.6 | — | — | — | |
| PMGBackbone=ResNet-502021.07 | 89.6 | — | — | — | |
| MMAL-NetBackbone=ResNet-502021.07 | 89.6 | — | — | — | |
| PMGBackbone=ResNet-502021.03 | 89.6 | — | — | — | |
| S3NBackbone=ResNet-50, Input in Testing=image2022.03 | 89.6 | — | — | — | |
| PMGPublication=ECCV 2020, Backbone=ResNet-502024.03 | 89.6 | — | — | — | |
| PARTPublication=TIP 2021, Backbone=ResNet-502024.03 | 89.6 | — | — | — | |
| Ours_NTSLabel Setting=Multi-granularity2020.11 | 89.53 | 96.57 | 91.58 | 80.45 | |
| DINOv2-Reg-BEvaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 89.52 | — | — | — | |
| CS-PartsBase Model=ResNet-502021.02 | 89.5 | — | — | — | |
| ME-ASNPublication=TMM 2022, Backbone=ResNet-502024.03 | 89.5 | — | — | — | |
| WS-DAN2020.10 | 89.4 | — | — | — | |
| DP-NetPublication=AAAI 2021, Backbone=ResNet-502024.03 | 89.3 | — | — | — | |
| PMGBackbone=ResNet50, Pre-training=iNaturalist2026.03 | 89.2 | — | — | — | |
| FDLBackbone=DenseNet-1612021.07 | 89.1 | — | — | — | |
| FDLBackbone=DenseNet-1612021.03 | 89.1 | — | — | — | |
| DPNPublication=TCSVT 2022, Backbone=ResNet-502024.03 | 89.1 | — | — | — | |
| DINOv2-BEvaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 89.06 | — | — | — | |
| PMGBase Model=ResNet502020.03 | 88.9 | — | — | — | |
| PCA-NetBackbone=ResNet-101, Input size=448 x 4482021.01 | 88.9 | — | — | — | |
| PF-CBM (spatial prior)Grounding=structural, Supervision=∼0.5% kp images2026.06 | 88.85 | — | — | — | |
| PF-CBM (box-free)Grounding=structural, Supervision=none2026.06 | 88.85 | — | — | — | |
| PMGBase Model=VGG16, Evaluation Mode=Combined Accuracy2020.03 | 88.8 | — | — | — | |
| DF-GMMBase Model=ResNet-502021.02 | 88.8 | — | — | — | |
| MPN-COV2020.10 | 88.7 | — | — | — | |
| RA-CNNBackbone=ResNet50, Pre-training=iNaturalist2026.03 | 88.7 | — | — | — | |
| Ours (multi scale)Backbone=ResNet-50, Resolution=448, #Parameters=23.9M, scale_mode=multi2019.12 | 88.6 | — | — | — | |
| API-NetBackbone=ResNet-101, Input size=448 x 4482021.01 | 88.6 | — | — | — | |
| MGE-CNNBase Model=ResNet502020.03 | 88.5 | — | — | — | |
| S3NBase Model=ResNet502020.03 | 88.5 | — | — | — | |
| MGEBase Model=ResNet-502021.02 | 88.5 | — | — | — | |
| S3NBackbone=ResNet-502021.07 | 88.5 | — | — | — | |
| S3NsPublication=ICCV 2019, Backbone=ResNet-502024.03 | 88.5 | — | — | — | |
| PMGBackbone=DenseNet161, Pre-training=ImageNet2026.03 | 88.5 | — | — | — | |
| AP-CNNPublication=TIP 2021, Backbone=ResNet-502024.03 | 88.4 | — | — | — | |
| AMPBackbone=ResNet50, Pre-training=iNaturalist2026.03 | 88.4 | — | — | — | |
| GCLBase Model=ResNet-502021.02 | 88.3 | — | — | — | |
| PCA-NetBackbone=ResNet-50, Input size=448 x 4482021.01 | 88.3 | — | — | — | |
| Ours_MCLabel Setting=Multi-granularity2020.11 | 88.26 | 96.58 | 90.36 | 77.85 | |
| OursLabel Setting=Multi-granularity2020.11 | 88.24 | 96.37 | 90.39 | 77.95 | |
| PMGBase Model=VGG162020.03 | 88.2 | — | — | — | |
| CMNPublication=TIP2022, Backbone=ResNet-502024.03 | 88.2 | — | — | — | |
| ACNetBase Model=ResNet-502021.02 | 88.1 | — | — | — | |
| CINBackbone=ResNet-1012021.07 | 88.1 | — | — | — | |
| DBTNetBackbone=ResNet-1012021.07 | 88.1 | — | — | — | |
| ACNetBackbone=ResNet-502021.07 | 88.1 | — | — | — | |
| DBTNetBackbone=ResNet-1012021.03 | 88.1 | — | — | — | |
| ACNetPublication=CVPR 2020, Backbone=ResNet-502024.03 | 88.1 | — | — | — | |
| ACNetBackbone=ResNet-50, Input size=448 x 4482021.01 | 88.1 | — | — | — | |
| CINBackbone=ResNet-101, Input size=448 x 4482021.01 | 88.1 | — | — | — | |
| VECA-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 256, Core budget (C)=642026.05 | 88.02 | — | — | — | |
| RA-CNNBackbone=DenseNet161, Pre-training=ImageNet2026.03 | 88 | — | — | — | |
| TASNBase Model=ResNet50, Model Component=D+A2020.02 | 87.9 | — | — | — | |
| TASNBackbone=ResNet-50, Input size=448 x 4482021.01 | 87.9 | — | — | — | |
| NTS-NetBackbone=ResNet50, Pre-training=iNaturalist2026.03 | 87.9 | — | — | — | |
| DCLBase Model=ResNet502020.03 | 87.8 | — | — | — | |
| DCLBase Model=ResNet-502021.02 | 87.8 | — | — | — |