Image Classification on VOC 2007 (test)
91.9mAPDINO (ViT-B/8)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DINO (ViT-B/8)Arch.=ViT-B/8, Pretrain=INet-1K, Param.=85M, Evaluation protocol=linear probe2022.02 | 91.9 | — | — | — | — | — | |
| SEER (RG-10B)Arch.=RG-10B, Pretrain=IG-1B, Param.=10B, Evaluation protocol=linear probe2022.02 | 91.8 | — | — | — | — | — | |
| SEER (RG-128Gf)Arch.=RG-128Gf, Pretrain=IG-1B, Param.=693M, Evaluation protocol=linear probe2022.02 | 91.6 | — | — | — | — | — | |
| SwAV (RG-128Gf)Arch.=RG-128Gf, Pretrain=INet-1K, Param.=693M, Evaluation protocol=linear probe2022.02 | 91.4 | — | — | — | — | — | |
| DINO (ViT-B/16)Arch.=ViT-B/16, Pretrain=INet-1K, Param.=85M, Evaluation protocol=linear probe2022.02 | 90.6 | — | — | — | — | — | |
| Supervised (ViT-B/16)Arch.=ViT-B/16, Pretrain=INet-1K, Param.=85M, Evaluation protocol=linear probe2022.02 | 90.5 | — | — | — | — | — | |
| SwAV (RN50w5)Arch.=RN50w5, Pretrain=INet-1K, Param.=585M, Evaluation protocol=linear probe2022.02 | 90.3 | — | — | — | — | — | |
| BYOLArch.=RN200w2, Pretrain=INet-1K, Param.=250M, Evaluation protocol=linear probe2022.02 | 90.1 | — | — | — | — | — | |
| VGG Net-D & Net-Emode=Late fusion of Net-D and Net-E2014.09 | 89.7 | — | — | — | — | — | |
| Supervised (RG-128Gf)Arch.=RG-128Gf, Pretrain=INet-1K, Param.=693M, Evaluation protocol=linear probe2022.02 | 89.4 | — | — | — | — | — | |
| MoCov3Arch.=ViT-B/16, Pretrain=INet-1K, Param.=85M, Evaluation protocol=linear probe2022.02 | 89.4 | — | — | — | — | — | |
| VGG Net-DNumber of layers=16 layers2014.09 | 89.3 | — | — | — | — | — | |
| VGG Net-ENumber of layers=19 layers2014.09 | 89.3 | — | — | — | — | — | |
| OBOWBackbone=ResNet-50, Evaluation Protocol=Linear Classification on frozen representations2023.03 | 89.3 | — | — | — | — | — | |
| SwAVEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 88.9 | — | — | — | — | — | |
| HEXADCluster⁺Epoch=200, Number of Crops=8-crop, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 88.8 | — | — | — | — | — | |
| VirTexPretrain Images=118K, Annotations=captions, Evaluation Protocol=Linear classification2020.06 | 88.7 | — | — | — | — | — | |
| DINO (RN50)Arch.=RN50, Pretrain=INet-1K, Param.=25M, Evaluation protocol=linear probe2022.02 | 88.5 | — | — | — | — | — | |
| SwAV (RN50)Arch.=RN50, Pretrain=INet-1K, Param.=25M, Evaluation protocol=linear probe2022.02 | 88.5 | — | — | — | — | — | |
| HEXADCluster⁺Epoch=800, Number of Crops=8-crop, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 88.2 | — | — | — | — | — | |
| SwAVEpoch=800, Batch size=4096, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 88.1 | — | — | — | — | — | |
| HEXADClusterEpoch=200, Number of Crops=8-crop, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 88.1 | — | — | — | — | — | |
| SwAVEpoch=200, Batch size=4096, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 87.9 | — | — | — | — | — | |
| HEXADClusterEpoch=800, Number of Crops=8-crop, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 87.9 | — | — | — | — | — | |
| SwAV (200 ep.)Pretrain Images=1.28M, Annotations=self-sup., Evaluation Protocol=Linear classification2020.06 | 87.9 | — | — | — | — | — | |
| DeepCluster-v2Epoch=800, Number of Crops=8-crop, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 87.6 | — | — | — | — | — | |
| SupervisedBackbone=ResNet-50, Parameters=25.6M, Evaluation Setup=[19]2019.12 | 87.5 | — | — | — | — | — | |
| SupervisedEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 87.5 | — | — | — | — | — | |
| SupervisedBackbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 87.5 | — | — | — | — | — | |
| SwAVEpoch=200, Batch size=256, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 87.5 | — | — | — | — | — | |
| ICMLMatt-fcPretrain Images=118K, Annotations=captions, Evaluation Protocol=Linear classification, Backbone Note=Uses pretrained BERT-base2020.06 | 87.5 | — | — | — | — | — | |
| SupervisedBackbone=ResNet-50, Evaluation Protocol=Linear Classification on frozen representations2023.03 | 87.5 | — | — | — | — | — | |
| MoCov2Evaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 87.1 | — | — | — | — | — | |
| MV-MRBackbone=ResNet-50, Evaluation Protocol=Linear Classification on frozen representations2023.03 | 87.1 | — | — | — | — | — | |
| HEXAMoCoEpoch=800, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 87 | — | — | — | — | — | |
| MoCo-v2Epoch=800, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 86.8 | — | — | — | — | — | |
| BYOLBackbone=ResNet-50, Evaluation Protocol=Linear Classification on frozen representations2023.03 | 86.6 | — | — | — | — | — | |
| VICRegBackbone=ResNet-50, Evaluation Protocol=Linear Classification on frozen representations2023.03 | 86.6 | — | — | — | — | — | |
| SimCLREvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 86.4 | — | — | — | — | — | |
| Barlow TwinsBackbone=ResNet-50, Evaluation Protocol=Linear Classification on frozen representations2023.03 | 86.2 | — | — | — | — | — | |
| HEXADClusterEpoch=200, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 85.9 | — | — | — | — | — | |
| PCL-v2Epoch=200, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 85.4 | — | — | — | — | — | |
| DeepCluster-v2Epoch=200, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 85.4 | — | — | — | — | — | |
| Wei et al.Pre-training=extended ILSVRC (2000 classes)2014.09 | 85.2 | — | — | — | — | — | |
| HEXAMoCoEpoch=200, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 85 | — | — | — | — | — | |
| MoCo-v2Epoch=200, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 84.5 | — | — | — | — | — | |
| PCLEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 84 | — | — | — | — | — | |
| PCL v1Pretrain Images=1.28M, Annotations=self-sup., Evaluation Protocol=Linear classification2020.06 | 83.1 | — | — | — | — | — | |
| Chatfield et al.2014.09 | 82.4 | — | — | — | — | — | |
| He et al.2014.09 | 82.4 | — | — | — | — | — | |
| Wei et al.Pre-training=ILSVRC-10002014.09 | 81.5 | — | — | — | — | — | |
| PIRLBackbone=ResNet-50, Parameters=25.6M, Evaluation Setup=[19]2019.12 | 81.1 | — | — | — | — | — | |
| PIRLEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 81.1 | — | — | — | — | — | |
| PIRLEpoch=800, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 81.1 | — | — | — | — | — | |
| PIRLBackbone=ResNet-50, Evaluation Protocol=Linear Classification on frozen representations2023.03 | 81.1 | — | — | — | — | — | |
| SimCLREpoch=1000, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 80.5 | — | — | — | — | — | |
| MoCoEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 79.8 | — | — | — | — | — | |
| MoCo-IN v1Pretrain Images=1.28M, Annotations=self-sup., Evaluation Protocol=Linear classification2020.06 | 79.4 | — | — | — | — | — | |
| BoWNetEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 79.3 | — | — | — | — | — | |
| MoCoEpoch=200, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 79.2 | — | — | — | — | — | |
| NPID++Backbone=ResNet-50, Parameters=25.6M, Evaluation Setup=[19]2019.12 | 76.6 | — | — | — | — | — | |
| NPID++Evaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 76.6 | — | — | — | — | — | |
| RotNetEvaluation protocol=Linear classification, Backbone=ResNet-50, Features=Frozen, Pre-trained=ImageNet2020.06 | 64.6 | — | — | — | — | — | |
| JigsawBackbone=ResNet-50, Parameters=25.6M, Evaluation Setup=[19]2019.12 | 64.5 | — | — | — | — | — | |
| JigsawEpoch=90, Backbone=ResNet-50, Evaluation Protocol=Linear classification2020.12 | 64.5 | — | — | — | — | — | |
| RotationBackbone=ResNet-50, Parameters=25.6M, Evaluation Setup=[19]2019.12 | 63.9 | — | — | — | — | — | |
| ColorizationBackbone=ResNet-50, Parameters=25.6M, Evaluation Setup=[19]2019.12 | 55.6 | — | — | — | — | — | |
| + HEXA_DCluster (8-crop)Epoch=200, Number of crops=8, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 56.9 | 67.3 | 76.4 | 81.1 | 84 | |
| + HEXA_DCluster (8-crop)Epoch=800, Number of crops=8, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 54.8 | 65.8 | 74.2 | 79.4 | 83.1 | |
| DeepCluster-v2Epoch=200, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 57.7 | 66.5 | 74.1 | 77.6 | 80.7 | |
| DeepCluster-v2Epoch=800, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 53.5 | 65.6 | 73.3 | 78.9 | 82.5 | |
| HEXA_DClusterEpoch=200, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 55.7 | 65.3 | 74.1 | 78 | 81.2 | |
| HEXA_DCluster (8-crop)Epoch=200, Number of crops=8, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 55.5 | 66.2 | 75.2 | 79.5 | 83.1 | |
| HEXA_DCluster (8-crop)Epoch=800, Number of crops=8, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 54.1 | 65.6 | 73.9 | 79 | 82.9 | |
| HEXA_MoCoEpoch=200, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 57 | 68.5 | 73.1 | 78 | 80.1 | |
| HEXA_MoCoEpoch=800, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 61.5 | 72.8 | 77.5 | 81.5 | 83 | |
| JigsawEpoch=200, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 31.1 | 40 | 46.7 | 51.8 | — | |
| MoCoEpoch=200, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 42 | 49.5 | 60 | 65.9 | — | |
| MoCo-v2Epoch=200, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 56.4 | 67.2 | 72 | 77.2 | 79.5 | |
| MoCo-v2Epoch=800, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 60.6 | 72.1 | 77.1 | 80.9 | 82.8 | |
| PCL-v2Epoch=200, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 59.6 | 66.2 | 74.5 | 78.3 | — | |
| SimCLREpoch=200, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 43.1 | 52.5 | 61 | 67.1 | — | |
| SupervisedEvaluation protocol=linear SVM on fixed representations2020.12 | — | 67.8 | 73.9 | 79.6 | 82.3 | 83.8 | |
| SwAV (B=4096)Epoch=200, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 53.8 | 65 | 73.9 | 78.6 | 82.3 | |
| SwAV (B=4096)Epoch=800, Evaluation protocol=linear SVM on fixed representations2020.12 | — | 54.6 | 64.6 | 73.4 | 79 | 82.5 |