Object Detection on VOC 07+12 (test)
79.3APallSupervised
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SupervisedPretrain Dataset=ImageNet-1M, Architecture=Resnet-50-FPN, Frozen conv body=true, Detector=Faster R-CNN2020.05 | 79.3 | — | — | — | |
| PCLPretrain Dataset=ImageNet-1M, Architecture=Resnet-50-FPN, Frozen conv body=true, Detector=Faster R-CNN2020.05 | 78.5 | — | — | — | |
| MoCoPretrain Dataset=ImageNet-1M, Architecture=Resnet-50-FPN, Frozen conv body=true, Detector=Faster R-CNN2020.05 | 73.5 | — | — | — | |
| MoCo-v2Architecture=Faster R-CNN, Backbone=C4 variant, Evaluation Protocol=finetuned2021.03 | 57.4 | 82.5 | 64 | — | |
| MoCo-v2transfer_learning=true2021.03 | 57.4 | 82.5 | 64 | — | |
| CW-RGPPre-training=Unsupervised ImageNet (200 epochs)2022.10 | 57.2 | 82.2 | 63.8 | — | |
| SimSiamArchitecture=Faster R-CNN, Backbone=C4 variant, Evaluation Protocol=finetuned2021.03 | 57 | 82.4 | 63.7 | — | |
| SimSiamtransfer_learning=true2021.03 | 57 | 82.4 | 63.7 | — | |
| MoCo v2Pre-training=Unsupervised ImageNet (200 epochs)2022.10 | 57 | 82.3 | 63.3 | — | |
| truncated tripletArchitecture=Faster-RCNN2021.04 | 56.9 | 82.6 | 63.8 | — | |
| Barlow TwinsArchitecture=Faster R-CNN, Backbone=C4 variant, Evaluation Protocol=finetuned2021.03 | 56.8 | 82.6 | 63.4 | — | |
| BarlowTtransfer_learning=true2021.03 | 56.8 | 82.6 | 63.4 | — | |
| MoCo v2Architecture=Faster-RCNN2021.04 | 56.6 | 82 | 62.9 | — | |
| Self-Classifiertransfer_learning=true2021.03 | 56.6 | 82.4 | 62.6 | — | |
| SimSiamPre-training=Unsupervised ImageNet (200 epochs)2022.10 | 56.4 | 82 | 62.8 | — | |
| SwAVArchitecture=Faster R-CNN, Backbone=C4 variant, Evaluation Protocol=finetuned2021.03 | 56.1 | 82.6 | 62.7 | — | |
| SwAVtransfer_learning=true2021.03 | 56.1 | 82.6 | 62.7 | — | |
| MoCoArchitecture=Faster-RCNN2021.04 | 56 | 81.4 | 62.2 | — | |
| Rotation-PredArchitecture=Faster-RCNN2021.04 | 55.5 | 80.9 | 61.4 | — | |
| SimCLRPre-training=Unsupervised ImageNet (200 epochs)2022.10 | 55.5 | 81.8 | 61.4 | — | |
| SwAVPre-training=Unsupervised ImageNet (200 epochs)2022.10 | 55.4 | 81.5 | 61.4 | — | |
| BYOLPre-training=Unsupervised ImageNet (200 epochs)2022.10 | 55.3 | 81.4 | 61.1 | — | |
| MoCoNetwork=ResNet-50, BatchNorm=Finetuned2019.12 | 55.2 | 81.4 | 61.2 | — | |
| Relative-LocArchitecture=Faster-RCNN2021.04 | 55.1 | 80.4 | 61.2 | — | |
| supervisedArchitecture=Faster-RCNN2021.04 | 54.2 | 81.6 | 59.8 | — | |
| NPIDArchitecture=Faster-RCNN2021.04 | 54.1 | 80 | 59.5 | — | |
| PIRLNetwork=ResNet-50, BatchNorm=Fixed2019.12 | 54 | 80.7 | 59.7 | 2.3 | |
| SupervisedArchitecture=Faster R-CNN, Backbone=C4 variant, Evaluation Protocol=finetuned2021.03 | 53.5 | 81.3 | 58.8 | — | |
| Supervisedtransfer_learning=true2021.03 | 53.5 | 81.3 | 58.8 | — | |
| IN-supervisedPre-training=Supervised ImageNet2022.10 | 53.5 | 81.3 | 58.8 | — | |
| SupervisedNetwork=ResNet-50, BatchNorm=Fixed2019.12 | 52.6 | 81.1 | 57.4 | 0 | |
| NPID++Network=ResNet-50, BatchNorm=Fixed2019.12 | 52.3 | 79.1 | 56.9 | -0.5 | |
| BYOLArchitecture=Faster-RCNN2021.04 | 51.9 | 81 | 56.5 | — | |
| SimCLRArchitecture=Faster-RCNN2021.04 | 51.5 | 79.4 | 55.6 | — | |
| JigsawNetwork=ResNet-50, BatchNorm=Fixed2019.12 | 48.9 | 75.1 | 52.9 | -4.5 | |
| RotationNetwork=ResNet-50, BatchNorm=Fixed2019.12 | 46.3 | 72.5 | 49.3 | -8.1 | |
| ScratchPre-training=None2022.10 | 33.8 | 60.2 | 33.1 | — | |
| RandomArchitecture=Faster-RCNN2021.04 | 32.8 | 59 | 31.6 | — | |
| CPC-BigNetwork=ResNet-101, BatchNorm=Fixed2019.12 | — | 70.6 | — | — | |
| CPC-HugeNetwork=ResNet-170, BatchNorm=Fixed2019.12 | — | 72.1 | — | — |