Fine-grained Classification on Food101
94.7Top-1 AccSigLIP2-B/16-256
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SigLIP2-B/16-256Evaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 94.7 | — | |
| DINOv3-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 94.47 | — | |
| DFN2B-CLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 93.32 | — | |
| EfficientNet B7Model=EfficientNet B72020.01 | 93 | — | |
| MP-FGVCBackbone=ViT-B-162023.09 | 93 | — | |
| DINOv2-Reg-BEvaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 92.87 | — | |
| Assemble-ResNet-FGVC-50Model=Assemble-ResNet-FGVC-502020.01 | 92.5 | — | |
| VECA-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 256, Core budget (C)=642026.05 | 92.44 | — | |
| CLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 92.42 | — | |
| TransFG*Backbone=ViT-B-162023.09 | 92.1 | — | |
| DINOv2-BEvaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 92.02 | — | |
| SIM-Trans*Backbone=ViT-B-162023.09 | 91.9 | — | |
| C-RADIOv3-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 91.64 | — | |
| IELT*Backbone=ViT-B-162023.09 | 91.4 | — | |
| OpenCLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 91.33 | — | |
| PRENetBackbone=CNN2023.09 | 91.1 | — | |
| MSMVFABackbone=CNN2023.09 | 90.6 | — | |
| IG-CMANBackbone=CNN2023.09 | 90.4 | — | |
| VECA-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 256, Core budget (C)=82026.05 | 90.04 | — | |
| SGLANetBackbone=CNN2023.09 | 89.7 | — | |
| DCLBackbone=CNN2023.09 | 88.9 | — | |
| ResNet-50Model=ResNet-502020.01 | 87 | — | |
| Fair Context Learning (FCL)mode=Test-Time Adaptation2026.02 | 86.57 | — | |
| CoOpPre-trained on ImageNet=true2024.10 | 85.3 | — | |
| TPTPre-trained on ImageNet=false2024.10 | 84.7 | — | |
| TPTmode=Test-Time Adaptation2026.02 | 84.66 | — | |
| MTAmode=Test-Time Adaptation2026.02 | 84.45 | — | |
| TPSmode=Test-Time Adaptation2026.02 | 84.4 | — | |
| R-TPTmode=Test-Time Adaptation2026.02 | 84.28 | — | |
| RLCFmode=Test-Time Adaptation2026.02 | 84.22 | — | |
| APMPre-trained on ImageNet=false2024.10 | 84.2 | — | |
| CoCoOpPre-trained on ImageNet=true2024.10 | 84 | — | |
| TTLmode=Test-Time Adaptation2026.02 | 83.99 | — | |
| ZEROmode=Test-Time Adaptation2026.02 | 83.78 | — | |
| CLIP-VIT-B/16Pre-trained on ImageNet=false2024.10 | 83.7 | — | |
| CLIPmode=Zero-shot2026.02 | 83.64 | — | |
| C-TPTmode=Test-Time Adaptation2026.02 | 83.17 | — | |
| EnsemblePre-trained on ImageNet=false2024.10 | 82.9 | — | |
| DiffTPTBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 79.21 | — | |
| DotaBackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Online2025.08 | 78.61 | — | |
| TransCLIPBackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Transductive2025.08 | 78 | — | |
| StatABackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Transductive2025.08 | 77.9 | — | |
| DPEBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 77.83 | — | |
| TDABackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Online2025.08 | 77.75 | — | |
| BCABackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Online2025.08 | 77.19 | — | |
| ADAPTBackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Transductive2025.08 | 76.97 | — | |
| DMNBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 76.7 | — | |
| TPSBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 76.2 | — | |
| ADAPTBackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Online2025.08 | 75.57 | — | |
| TPTBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 74.88 | — | |
| C-TPTBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 74.7 | — | |
| CLIPBackbone=CLIP-RN50, BP-free=N/A, Evaluation Protocol=Online2025.08 | 73.97 | — | |
| Bilinear-CNN2020.10 | — | 82.4 | |
| BITS-fixedProtocol=Linear probing, Weight initialization=Frozen ImageNet-1k pre-trained weights2026.02 | — | 81.52 | |
| BITS-fixedProtocol=Linear probing, Weight initialization=Self-supervised fine-tuned on target domain2026.02 | — | 82.88 | |
| BITS-resetProtocol=Linear probing, Weight initialization=Frozen ImageNet-1k pre-trained weights2026.02 | — | 74.78 | |
| BITS-resetProtocol=Linear probing, Weight initialization=Self-supervised fine-tuned on target domain2026.02 | — | 82.69 | |
| CAPTraining data setup=primary2021.01 | — | 98.6 | |
| CLIPPre-training Dataset=RedCaps, Pre-training Data Type=Real, Pre-training Schedule=35 epochs, Backbone=ViT-B/162023.06 | — | 89.2 | |
| CLIPPre-training Dataset=RedCaps, Pre-training Data Type=Syn, Pre-training Schedule=35 epochs, Backbone=ViT-B/162023.06 | — | 83.7 | |
| CLIPPre-training Dataset=WIT-400M, Pre-training Data Type=Real, Backbone=ViT-B/162023.06 | — | 92.8 | |
| DATL + WS-DAN2020.10 | — | 88.7 | |
| DINOProtocol=Linear probing, Weight initialization=Frozen ImageNet-1k pre-trained weights2026.02 | — | 75.59 | |
| DINOProtocol=Linear probing, Weight initialization=Self-supervised fine-tuned on target domain2026.02 | — | 70.9 | |
| DLA2020.10 | — | 89.7 | |
| DSTLTraining data setup=primary + secondary2021.01 | — | 90.4 | |
| GPipeTraining data setup=primary2021.01 | — | 93 | |
| JDNetTraining data setup=primary + secondary2021.01 | — | 91.2 | |
| MSMVFATraining data setup=primary2021.01 | — | 90.6 | |
| ResNet-50 (Baseline)Backbone=ResNet-50, Pre-training strategy=Standard2021.01 | — | 87.98 | |
| ResNet-50 (ReLabel-trained)Backbone=ResNet-50, Pre-training strategy=ReLabel2021.01 | — | 88.12 | |
| SimCLRPre-training Dataset=RedCaps, Pre-training Data Type=Real, Pre-training Schedule=35 epochs, Backbone=ViT-B/162023.06 | — | 81.3 | |
| SimCLRPre-training Dataset=RedCaps, Pre-training Data Type=Syn, Pre-training Schedule=35 epochs, Backbone=ViT-B/162023.06 | — | 79.7 | |
| SimDINOProtocol=Linear probing, Weight initialization=Frozen ImageNet-1k pre-trained weights2026.02 | — | 75.79 | |
| SimDINOProtocol=Linear probing, Weight initialization=Self-supervised fine-tuned on target domain2026.02 | — | 66.19 | |
| StableRepPre-training Dataset=RedCaps, Pre-training Data Type=Syn, Pre-training Schedule=35 epochs, Backbone=ViT-B/162023.06 | — | 88.1 | |
| StableRepPre-training Dataset=CC12M, Pre-training Data Type=Syn, Pre-training Schedule=35 epochs, Backbone=ViT-B/162023.06 | — | 83.6 | |
| StableRepPre-training Dataset=CC12M, Pre-training Data Type=Syn, Pre-training Schedule=105 epochs, Backbone=ViT-B/162023.06 | — | 85.1 | |
| StableRepPre-training Dataset=RedCaps, Pre-training Data Type=Syn, Pre-training Schedule=105 epochs, Backbone=ViT-B/162023.06 | — | 89.4 | |
| Subset B2020.10 | — | 90.4 | |
| WISERTraining data setup=primary2021.01 | — | 90.3 | |
| WS-DAN2020.10 | — | 87.2 |