Image Classification on ImageNet (val) (Accuracy)
91AccuracyCoCa
Evaluation results
| Method | Links | |
|---|---|---|
| CoCaStatus=Fine-tuned SOTA2022.06 | 91 | |
| CoCaEvaluation Protocol=Zero-shot2023.02 | 86.3 | |
| LiT-22BEvaluation Protocol=Zero-shot, Backbone=ViT-22B/142023.02 | 85.9 | |
| BASICEvaluation Protocol=Zero-shot2023.02 | 85.7 | |
| LiT-e/14Evaluation Protocol=Zero-shot, Backbone=ViT-e/142023.02 | 85.4 | |
| LiT-g/14Evaluation Protocol=Zero-shot, Backbone=ViT-g/142023.02 | 85.2 | |
| MAE-CTaugArchitecture=ViT-H/16, Feature Evaluation Protocol=Linear probing2023.04 | 82.2 | |
| MAE-CTaugArchitecture=ViT-L/16, Feature Evaluation Protocol=Linear probing2023.04 | 81.5 | |
| MAE-CTminArchitecture=ViT-H/16, Feature Evaluation Protocol=Linear probing2023.04 | 81.5 | |
| iBOTArchitecture=ViT-L/16, Feature Evaluation Protocol=Linear probing2023.04 | 81 | |
| Layer GraftingArchitecture=ViT-L/16, Feature Evaluation Protocol=Linear probing2023.04 | 81 | |
| MAE-CTminArchitecture=ViT-L/16, Feature Evaluation Protocol=Linear probing2023.04 | 80.2 | |
| MAE-CTaugArchitecture=ViT-H/16, Feature Evaluation Protocol=k-NN2023.04 | 79.8 | |
| iBOTArchitecture=ViT-B/16, Feature Evaluation Protocol=Linear probing2023.04 | 79.5 | |
| MAE-CTminArchitecture=ViT-H/16, Feature Evaluation Protocol=k-NN2023.04 | 79.4 | |
| UNIFIED-IO XLModel Size=XL2022.06 | 79.1 | |
| MAE-CTaugArchitecture=ViT-L/16, Feature Evaluation Protocol=k-NN2023.04 | 79.1 | |
| MAEArchitecture=ViT-H/16, Feature Evaluation Protocol=Linear probing2023.04 | 78 | |
| iBOTArchitecture=ViT-L/16, Feature Evaluation Protocol=k-NN2023.04 | 78 | |
| MAE-CTminArchitecture=ViT-L/16, Feature Evaluation Protocol=k-NN2023.04 | 78 | |
| MSNArchitecture=ViT-B/16, Feature Evaluation Protocol=Linear probing2023.04 | 77.7 | |
| Layer GraftingArchitecture=ViT-B/16, Feature Evaluation Protocol=Linear probing2023.04 | 77.7 | |
| MSNArchitecture=ViT-L/16, Feature Evaluation Protocol=Linear probing2023.04 | 77.3 | |
| Layer GraftingArchitecture=ViT-L/16, Feature Evaluation Protocol=k-NN2023.04 | 77.3 | |
| iBOTArchitecture=ViT-B/16, Feature Evaluation Protocol=k-NN2023.04 | 77.1 | |
| MAE-CTaugArchitecture=ViT-B/16, Feature Evaluation Protocol=Linear probing2023.04 | 76.9 | |
| MoCo v3Architecture=ViT-B/16, Feature Evaluation Protocol=Linear probing2023.04 | 76.7 | |
| ALIGNEvaluation Protocol=Zero-shot2023.02 | 76.4 | |
| MSNArchitecture=ViT-B/16, Feature Evaluation Protocol=k-NN2023.04 | 76.3 | |
| CLIPEvaluation Protocol=Zero-shot2023.02 | 76.2 | |
| MSNArchitecture=ViT-L/16, Feature Evaluation Protocol=k-NN2023.04 | 76.2 | |
| Train from ScratchBackbone=ResNet-50, Model Size (MB)=5542019.10 | 76.16 | |
| FinetuneBackbone=ResNet-50, Model Size (MB)=5542019.10 | 76.16 | |
| ProgressiveNetBackbone=ResNet-50, Model Size (MB)=5632019.10 | 76.16 | |
| PiggybackBackbone=ResNet-50, Model Size (MB)=1212019.10 | 76.16 | |
| MAEArchitecture=ViT-L/16, Feature Evaluation Protocol=Linear probing2023.04 | 76 | |
| CPGBackbone=ResNet-50, Model Size (MB)=1212019.10 | 75.81 | |
| PackNetBackbone=ResNet-50, Model Size (MB)=1152019.10 | 75.71 | |
| Layer GraftingArchitecture=ViT-B/16, Feature Evaluation Protocol=k-NN2023.04 | 75.4 | |
| UniQBackbone=ResNet-34, Bit-width (W/A)=4/42021.04 | 75 | |
| QKDBackbone=ResNet-34, Bit-width (W/A)=4/42021.04 | 74.6 | |
| UniQBackbone=ResNet-34, Bit-width (W/A)=3/32021.04 | 74.2 | |
| LSQBackbone=ResNet-34, Bit-width (W/A)=4/42021.04 | 74.1 | |
| QKDBackbone=ResNet-34, Bit-width (W/A)=3/32021.04 | 73.9 | |
| QILBackbone=ResNet-34, Bit-width (W/A)=4/42021.04 | 73.7 | |
| MAE-CTminArchitecture=ViT-B/16, Feature Evaluation Protocol=Linear probing2023.04 | 73.5 | |
| LSQBackbone=ResNet-34, Bit-width (W/A)=3/32021.04 | 73.4 | |
| MAE-CTaugArchitecture=ViT-B/16, Feature Evaluation Protocol=k-NN2023.04 | 73.4 | |
| QILBackbone=ResNet-34, Bit-width (W/A)=3/32021.04 | 73.1 | |
| DSQBackbone=ResNet-34, Bit-width (W/A)=4/42021.04 | 72.8 | |
| MoCo v3Architecture=ViT-B/16, Feature Evaluation Protocol=k-NN2023.04 | 72.6 | |
| DSQBackbone=ResNet-34, Bit-width (W/A)=3/32021.04 | 72.5 | |
| UniQBackbone=ResNet-34, Bit-width (W/A)=2/22021.04 | 72.1 | |
| UNIFIED-IO LARGEModel Size=LARGE2022.06 | 71.8 | |
| LSQBackbone=ResNet-34, Bit-width (W/A)=2/22021.04 | 71.6 | |
| QKDBackbone=ResNet-34, Bit-width (W/A)=2/22021.04 | 71.6 | |
| CoOp2022.10 | 71.51 | |
| UniQBackbone=ResNet-18, Bit-width (W/A)=4/42021.04 | 71.5 | |
| QKDBackbone=ResNet-18, Bit-width (W/A)=4/42021.04 | 71.4 | |
| LSQBackbone=ResNet-18, Bit-width (W/A)=4/42021.04 | 71.1 | |
| CoCoOp2022.10 | 71.02 | |
| LSQ+Backbone=ResNet-18, Bit-width (W/A)=4/42021.04 | 70.8 | |
| Bayesian Prompt Learning2022.10 | 70.7 | |
| QILBackbone=ResNet-34, Bit-width (W/A)=2/22021.04 | 70.6 | |
| UniQBackbone=ResNet-18, Bit-width (W/A)=3/32021.04 | 70.5 | |
| SATBackbone=ResNet-18, Bit-width (W/A)=4/42021.04 | 70.3 | |
| LSQBackbone=ResNet-18, Bit-width (W/A)=3/32021.04 | 70.2 | |
| QKDBackbone=ResNet-18, Bit-width (W/A)=3/32021.04 | 70.2 | |
| QILBackbone=ResNet-18, Bit-width (W/A)=4/42021.04 | 70.1 | |
| DSQBackbone=ResNet-34, Bit-width (W/A)=2/22021.04 | 70 | |
| DSQBackbone=ResNet-18, Bit-width (W/A)=4/42021.04 | 69.6 | |
| LSQ+Backbone=ResNet-18, Bit-width (W/A)=3/32021.04 | 69.3 | |
| SATBackbone=ResNet-18, Bit-width (W/A)=3/32021.04 | 69.3 | |
| PACTBackbone=ResNet-18, Bit-width (W/A)=4/42021.04 | 69.2 | |
| QILBackbone=ResNet-18, Bit-width (W/A)=3/32021.04 | 69.2 | |
| DSQBackbone=ResNet-18, Bit-width (W/A)=3/32021.04 | 68.7 | |
| UniQBackbone=MobileNet-V2, Bit-width (W/A)=4/42021.04 | 68.2 | |
| PACTBackbone=ResNet-18, Bit-width (W/A)=3/32021.04 | 68.1 | |
| DoReFa-NetBackbone=ResNet-18, Bit-width (W/A)=4/42021.04 | 68.1 | |
| MAEArchitecture=ViT-B/16, Feature Evaluation Protocol=Linear probing2023.04 | 68 | |
| UniQBackbone=ResNet-18, Bit-width (W/A)=2/22021.04 | 67.8 | |
| LSQBackbone=ResNet-18, Bit-width (W/A)=2/22021.04 | 67.6 | |
| DoReFa-NetBackbone=ResNet-18, Bit-width (W/A)=3/32021.04 | 67.5 | |
| QKDBackbone=ResNet-18, Bit-width (W/A)=2/22021.04 | 67.4 | |
| QKDBackbone=MobileNet-V2, Bit-width (W/A)=4/42021.04 | 67.4 | |
| LSQ+Backbone=ResNet-18, Bit-width (W/A)=2/22021.04 | 66.8 | |
| UniQBackbone=ResNet-34, Bit-width (W/A)=1/12021.04 | 65.8 | |
| QILBackbone=ResNet-18, Bit-width (W/A)=2/22021.04 | 65.7 | |
| SATBackbone=ResNet-18, Bit-width (W/A)=2/22021.04 | 65.5 | |
| DSQBackbone=ResNet-18, Bit-width (W/A)=2/22021.04 | 65.2 | |
| UniQBackbone=MobileNet-V2, Bit-width (W/A)=3/32021.04 | 65 | |
| QILBackbone=MobileNet-V2, Bit-width (W/A)=4/42021.04 | 64.8 | |
| PACTBackbone=ResNet-18, Bit-width (W/A)=2/22021.04 | 64.4 | |
| MAE-CTminArchitecture=ViT-B/16, Feature Evaluation Protocol=k-NN2023.04 | 64.1 | |
| UNIFIED-IO BASEModel Size=BASE2022.06 | 63.3 | |
| OTTERBackbone=ViT-B/16, Setting=Zero-shot2024.04 | 62.9 | |
| DoReFa-NetBackbone=ResNet-18, Bit-width (W/A)=2/22021.04 | 62.6 | |
| QKDBackbone=MobileNet-V2, Bit-width (W/A)=3/32021.04 | 62.6 | |
| PACTBackbone=MobileNet-V2, Bit-width (W/A)=4/42021.04 | 61.4 | |
| MAEArchitecture=ViT-H/16, Feature Evaluation Protocol=k-NN2023.04 | 61.1 |