Fine-grained Image Classification on NABirds
91.45AccuracyTRD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TRD2023.10 | 91.45 | — | |
| Relational Proxies2023.10 | 91.2 | — | |
| CAPTraining data setup=primary2021.01 | 91 | — | |
| CAP2023.10 | 91 | — | |
| TransFG2023.10 | 90.8 | — | |
| FFVT2023.10 | 89.42 | — | |
| GaRD2023.10 | 88 | — | |
| DSTLTraining data setup=primary + secondary2021.01 | 87.9 | — | |
| MMAL2023.10 | 87.1 | — | |
| Cross-XTraining data setup=primary2021.01 | 86.4 | — | |
| MaxEntTraining data setup=primary + secondary2021.01 | 83 | — | |
| PC-CNNTraining data setup=primary2021.01 | 82.8 | — | |
| T-LossTraining data setup=primary2021.01 | 79.6 | — | |
| StochNorm2023.10 | 74.94 | — | |
| Cross X (SENet)Backbone=SENet-50, 1-Stage=true, Sep. Init.=false2019.09 | 0.864 | — | |
| Cross X (ResNet)Backbone=ResNet-50, 1-Stage=true, Sep. Init.=false2019.09 | 0.862 | — | |
| MaxEnt-CNNBackbone=DenseNet-161, 1-Stage=true, Sep. Init.=false2019.09 | 0.83 | — | |
| ResNet-501-Stage=true, Sep. Init.=false2019.09 | 0.822 | — | |
| SENet-501-Stage=true, Sep. Init.=false2019.09 | 0.821 | — | |
| PN-CNN1-Stage=false, Sep. Init.=true2019.09 | 0.74 | — | |
| MaxEnt-CNNBackbone=ResNet-50, 1-Stage=true, Sep. Init.=false2019.09 | 0.692 | — | |
| AlexNet-fc61-Stage=false, Sep. Init.=true2019.09 | 0.35 | — | |
| BIOCAPZero-shot=true, Visual Encoder=ViT-B/162025.10 | — | 67.6 | |
| BIOCLIPZero-shot=true, Visual Encoder=ViT-B/162025.10 | — | 58.8 | |
| BioTrove-CLIPZero-shot=true, Visual Encoder=ViT-B/162025.10 | — | 39.4 | |
| BYOLBackbone=ResNet-50, Pre-training Setting=Second-stage only (800 epochs), Evaluation Protocol=Fine-tuned2021.06 | — | 72.54 | |
| BYOL+CVSABackbone=ResNet-50, Pre-training Setting=Second-stage only (800 epochs), Evaluation Protocol=Fine-tuned2021.06 | — | 73.75 | |
| BYOL+DiLoBackbone=ResNet-50, Pre-training Setting=Second-stage only (800 epochs), Evaluation Protocol=Fine-tuned2021.06 | — | 73.12 | |
| CLIP (ViT-B/16)Zero-shot=true, Visual Encoder=ViT-B/162025.10 | — | 39 | |
| FG-CLIPZero-shot=true, Visual Encoder=ViT-B/162025.10 | — | 48.3 | |
| InsLocBackbone=ResNet-50, Pre-training Setting=Second-stage only (800 epochs), Evaluation Protocol=Fine-tuned2021.06 | — | 72.8 | |
| LooC*Backbone=ResNet-50, Pre-training Setting=Second-stage only (800 epochs), Evaluation Protocol=Fine-tuned2021.06 | — | 72.84 | |
| MoCo.v2Backbone=ResNet-50, Pre-training Setting=Second-stage only (800 epochs), Evaluation Protocol=Fine-tuned2021.06 | — | 71.36 | |
| RandomBackbone=ResNet-50, Pre-training Setting=Random initialization, Evaluation Protocol=Fine-tuned2021.06 | — | 65.78 | |
| Random GuessingZero-shot=true2025.10 | — | 0.2 | |
| Rel-LocBackbone=ResNet-50, Pre-training Setting=Second-stage only (800 epochs), Evaluation Protocol=Fine-tuned2021.06 | — | 72.6 | |
| Rot-PredBackbone=ResNet-50, Pre-training Setting=Second-stage only (800 epochs), Evaluation Protocol=Fine-tuned2021.06 | — | 73.01 | |
| SigLIPZero-shot=true, Visual Encoder=ViT-B/162025.10 | — | 50.2 | |
| SimCLRBackbone=ResNet-50, Pre-training Setting=Second-stage only (800 epochs), Evaluation Protocol=Fine-tuned2021.06 | — | 72.05 |