Object Detection on PASCAL VOC 2007+2012 (test)
85mAP (mean Average Precision)Faster RCNN
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Faster RCNN2018.12 | 85 | 82.1 | 82.3 | 86.5 | 89.3 | 85.6 | 84.3 | — | — | — | — | |
| Oracle2018.12 | 85 | 82.1 | 82.3 | 86.5 | 89.3 | 85.6 | 84.3 | — | — | — | — | |
| Strong-Weak Domain AlignmentGlobal Alignment (G)=true, Instance-level Alignment (I)=true2018.12 | 84.9 | 79.4 | 84.6 | 85.8 | 89.6 | 85.7 | 84.1 | — | — | — | — | |
| Strong-Weak Domain AlignmentGlobal Alignment (G)=true, Instance-level Alignment (I)=true, Context-level Alignment (CTX)=true, Local Alignment (L)=true2018.12 | 84.9 | 79.8 | 87.4 | 85.5 | 88.1 | 84.5 | 84 | — | — | — | — | |
| LeRaCModel=YOLOv52022.05 | 84.6 | — | — | — | — | — | — | — | — | — | — | |
| Strong-Weak Domain AlignmentGlobal Alignment (G)=true, Instance-level Alignment (I)=true, Local Alignment (L)=true2018.12 | 84.4 | 79.5 | 84.6 | 86.1 | 89.2 | 84.5 | 82.8 | — | — | — | — | |
| Faster RCNN*CycleGAN (P)=true2018.12 | 84.1 | 79.7 | 82.7 | 86 | 88.9 | 85 | 82.1 | — | — | — | — | |
| BDC-Faster2018.12 | 84 | 80.9 | 82.6 | 86.4 | 87.9 | 82.7 | 83.3 | — | — | — | — | |
| Strong-Weak Domain AlignmentGlobal Alignment (G)=true, Instance-level Alignment (I)=true, Context-level Alignment (CTX)=true, Local Alignment (L)=true, CycleGAN (P)=true2018.12 | 83.7 | 78.9 | 83.1 | 84.2 | 87.4 | 85.6 | 82.8 | — | — | — | — | |
| conventionalModel=YOLOv52022.05 | 83.2 | — | — | — | — | — | — | — | — | — | — | |
| DA-FasterGlobal Alignment (G)=true, Instance-level Alignment (I)=true2018.12 | 83.1 | 75.7 | 84.4 | 84.5 | 88.5 | 83.6 | 81.6 | — | — | — | — | |
| Strong-Weak Domain AlignmentGlobal Alignment (G)=true, Instance-level Alignment (I)=true, Context-level Alignment (CTX)=true2018.12 | 83.1 | 80 | 82 | 84.7 | 86.8 | 83.6 | 81.3 | — | — | — | — | |
| CBSModel=YOLOv52022.05 | 82.9 | — | — | — | — | — | — | — | — | — | — | |
| SSD512-RN50Backbone=ResNet-50, Input Resolution=512x512, Pruning Strategy=base model2022.10 | 77.98 | — | — | — | — | — | — | — | — | 65.56 | 68.24 | |
| SSD512-RN50-HALPBackbone=ResNet-50, Input Resolution=512x512, Pruning Strategy=HALP2022.10 | 77.42 | — | — | — | — | — | — | — | — | 15.38 | 132.57 | |
| RetinaNet-RN50Backbone=ResNet-50, Architecture=RetinaNet2022.10 | 77.27 | — | — | — | — | — | — | — | — | 106.5 | 36.92 | |
| SSD300-VGG16Backbone=VGG16, Input Resolution=300x3002022.10 | 76.72 | — | — | — | — | — | — | — | — | 31.44 | 122.28 | |
| SSD512-RN50-slimBackbone=ResNet-50, Input Resolution=512x512, Pruning Strategy=slimming2022.10 | 75.83 | — | — | — | — | — | — | — | — | 46.09 | 76.49 | |
| SSD300-RN50Backbone=ResNet-50, Input Resolution=300x3002022.10 | 75.69 | — | — | — | — | — | — | — | — | 16.23 | 128.85 | |
| FasterRCNN-VGG16Backbone=VGG16, Architecture=Faster R-CNN2022.10 | 70.1 | — | — | — | — | — | — | — | — | 91.23 | 29.21 | |
| DETRegDetector=Deformable DETR2021.06 | 63.5 | — | — | — | — | — | — | 83.3 | 70.3 | — | — | |
| YOLO-Master-NLatency (ms)=1.622025.12 | 62.1 | — | — | — | — | — | — | 81.9 | — | — | — | |
| SwAVDetector=Deformable DETR2021.06 | 61 | — | — | — | — | — | — | 83 | 68.1 | — | — | |
| YOLOv11-NLatency (ms)=1.52025.12 | 61 | — | — | — | — | — | — | 81.2 | — | — | — | |
| YOLOv12-NLatency (ms)=1.642025.12 | 60.7 | — | — | — | — | — | — | 80.8 | — | — | — | |
| YOLOv13-NLatency (ms)=1.972025.12 | 60.7 | — | — | — | — | — | — | 80.3 | — | — | — | |
| YOLOv10-NLatency (ms)=1.842025.12 | 60.6 | — | — | — | — | — | — | 80.3 | — | — | — | |
| SupervisedDetector=Deformable DETR2021.06 | 59.5 | — | — | — | — | — | — | 82.6 | 65.6 | — | — | |
| SoCoBackbone=ResNet-50-C4, Epochs=1002024.10 | 59.1 | — | — | — | — | — | — | 83.4 | 65.6 | — | — | |
| IQDet2021.04 | 59 | — | — | — | — | — | — | 81.4 | 65 | — | — | |
| ReSimBackbone=ResNet-50-C42024.10 | 58.7 | — | — | — | — | — | — | 83.1 | 66.3 | — | — | |
| InsLocEpoch=4002021.02 | 58.4 | — | — | — | — | — | — | 83 | 65.3 | — | — | |
| PAA(w/o vote)2021.04 | 58.3 | — | — | — | — | — | — | 80.7 | 64.4 | — | — | |
| DetCoArchitecture=Faster R-CNN R50-C4, Pre-train data=ImageNet, Epoch=800, Protocol=Fine-tuned2022.03 | 58.2 | — | — | — | — | — | — | 82.7 | 65 | — | — | |
| UniVIPArchitecture=Faster R-CNN R50-C4, Pre-train data=COCO+, Epoch=800, Protocol=Fine-tuned2022.03 | 58.2 | — | — | — | — | — | — | 83.3 | 65.2 | — | — | |
| InsLocEpoch=2002021.02 | 57.9 | — | — | — | — | — | — | 82.9 | 64.9 | — | — | |
| InsLocBackbone=ResNet-50-C42024.10 | 57.9 | — | — | — | — | — | — | 82.9 | 64.9 | — | — | |
| DetCoArchitecture=Faster R-CNN R50-C4, Pre-train data=ImageNet, Epoch=200, Protocol=Fine-tuned2022.03 | 57.8 | — | — | — | — | — | — | 82.6 | 64.2 | — | — | |
| DetCoBackbone=ResNet-50-C42024.10 | 57.8 | — | — | — | — | — | — | 82.6 | 64.2 | — | — | |
| InfoMinEpoch=2002021.02 | 57.6 | — | — | — | — | — | — | 82.7 | 64.6 | — | — | |
| InfoMinBackbone=ResNet-50-C42024.10 | 57.6 | — | — | — | — | — | — | 82.7 | 64.6 | — | — | |
| InfoMinEpoch=8002021.02 | 57.5 | — | — | — | — | — | — | 82.7 | 64.3 | — | — | |
| MoCHIBackbone=ResNet-50-C42024.10 | 57.5 | — | — | — | — | — | — | 82.7 | 64.4 | — | — | |
| MoCo-v2Epoch=8002021.02 | 57.4 | — | — | — | — | — | — | 82.5 | 64 | — | — | |
| MoCov2Architecture=Faster R-CNN R50-C4, Pre-train data=ImageNet, Epoch=800, Protocol=Fine-tuned2022.03 | 57.4 | — | — | — | — | — | — | 82.5 | 64 | — | — | |
| LEWEL_MEpochs=200, Backbone=ResNet-50 C4, Architecture=Faster R-CNN2022.03 | 57.3 | — | — | — | — | — | — | 82.3 | 63.6 | — | — | |
| Ourspretrain=Ours (BYOL + STS)2021.10 | 57.2 | — | — | — | — | — | — | 82.3 | 64.4 | — | — | |
| SynCoBackbone=ResNet-50-C42024.10 | 57.2 | — | — | — | — | — | — | 82.6 | 63.9 | — | — | |
| MoCo-v2Epoch=2002021.02 | 57 | — | — | — | — | — | — | 82.4 | 63.6 | — | — | |
| MoCo v2pretrain=MoCo v22021.10 | 57 | — | — | — | — | — | — | 82.3 | 63.3 | — | — | |
| SimSiampretrain=SimSiam2021.10 | 57 | — | — | — | — | — | — | 82.4 | 63.7 | — | — | |
| MoCov2Epochs=200, Backbone=ResNet-50 C4, Architecture=Faster R-CNN, reproduction=true2022.03 | 57 | — | — | — | — | — | — | 82.2 | 63.4 | — | — | |
| SimSiamBackbone=ResNet-50-C42024.10 | 57 | — | — | — | — | — | — | 82.4 | 63.7 | — | — | |
| MoCo-v2Backbone=ResNet-50-C42024.10 | 57 | — | — | — | — | — | — | 82.4 | 63.6 | — | — | |
| BTBackbone=ResNet-50-C4, Epochs=10002024.10 | 56.8 | — | — | — | — | — | — | 82.6 | 63.4 | — | — | |
| ATSS2021.04 | 56.7 | — | — | — | — | — | — | 79.9 | 62.3 | — | — | |
| UniVIPArchitecture=Faster R-CNN R50-C4, Pre-train data=COCO, Epoch=800, Protocol=Fine-tuned2022.03 | 56.5 | — | — | — | — | — | — | 82.3 | 62.6 | — | — | |
| LEWEL_MEpochs=100, Backbone=ResNet-50 C4, Architecture=Faster R-CNN2022.03 | 56.5 | — | — | — | — | — | — | 82.1 | 63 | — | — | |
| LEWEL_BEpochs=200, Backbone=ResNet-50 C4, Architecture=Faster R-CNN2022.03 | 56.5 | — | — | — | — | — | — | 82.6 | 63.7 | — | — | |
| BYOLArchitecture=Faster R-CNN R50-C4, Pre-train data=COCO+, Epoch=800, Protocol=Fine-tuned2022.03 | 56.4 | — | — | — | — | — | — | 81.9 | 62.6 | — | — | |
| SimSiamEpochs=200, Backbone=ResNet-50 C4, Architecture=Faster R-CNN, source=cited from [9]2022.03 | 56.4 | — | — | — | — | — | — | 82 | 62.8 | — | — | |
| SimCLRBackbone=ResNet-50-C4, Epochs=10002024.10 | 56.3 | — | — | — | — | — | — | 81.9 | 62.5 | — | — | |
| FCOS*2021.04 | 56.2 | — | — | — | — | — | — | 79.9 | 61.9 | — | — | |
| SwAVArchitecture=Faster R-CNN R50-C4, Pre-train data=ImageNet, Epoch=800, Protocol=Fine-tuned2022.03 | 56.1 | — | — | — | — | — | — | 82.6 | 62.7 | — | — | |
| MoCov2Epochs=100, Backbone=ResNet-50 C4, Architecture=Faster R-CNN, reproduction=true2022.03 | 56.1 | — | — | — | — | — | — | 81.5 | 62.4 | — | — | |
| LEWEL_BEpochs=100, Backbone=ResNet-50 C4, Architecture=Faster R-CNN2022.03 | 56.1 | — | — | — | — | — | — | 82.1 | 62.3 | — | — | |
| SwAVBackbone=ResNet-50-C4, Epochs=8002024.10 | 56.1 | — | — | — | — | — | — | 82.6 | 62.7 | — | — | |
| MoCoArchitecture=Faster R-CNN R50-C4, Pre-train data=ImageNet, Epoch=200, Protocol=Fine-tuned2022.03 | 55.9 | — | — | — | — | — | — | 81.5 | 62.6 | — | — | |
| MoCoBackbone=ResNet-50-C42024.10 | 55.9 | — | — | — | — | — | — | 81.5 | 62.6 | — | — | |
| BYOLEpochs=200, Backbone=ResNet-50 C4, Architecture=Faster R-CNN, reproduction=true2022.03 | 55.8 | — | — | — | — | — | — | 81.6 | 61.6 | — | — | |
| SimCLRpretrain=SimCLR2021.10 | 55.5 | — | — | — | — | — | — | 81.8 | 58.8 | — | — | |
| PIRLArchitecture=Faster R-CNN R50-C4, Pre-train data=ImageNet, Epoch=200, Protocol=Fine-tuned2022.03 | 55.5 | — | — | — | — | — | — | 81 | 61.3 | — | — | |
| BYOLEpochs=100, Backbone=ResNet-50 C4, Architecture=Faster R-CNN, reproduction=true2022.03 | 55.5 | — | — | — | — | — | — | 81.9 | 61.2 | — | — | |
| SimCLREpochs=200, Backbone=ResNet-50 C4, Architecture=Faster R-CNN, source=cited from [9]2022.03 | 55.5 | — | — | — | — | — | — | 81.8 | 61.4 | — | — | |
| PIRLBackbone=ResNet-50-C42024.10 | 55.5 | — | — | — | — | — | — | 81 | 61.3 | — | — | |
| SwAVpretrain=SwAV2021.10 | 55.4 | — | — | — | — | — | — | 81.5 | 61.4 | — | — | |
| SwAVEpochs=200, Backbone=ResNet-50 C4, Architecture=Faster R-CNN, source=cited from [9]2022.03 | 55.4 | — | — | — | — | — | — | 81.5 | 61.4 | — | — | |
| BYOLpretrain=BYOL2021.10 | 55.3 | — | — | — | — | — | — | 81.4 | 61.1 | — | — | |
| BYOLEpochs=200, Backbone=ResNet-50 C4, Architecture=Faster R-CNN, source=cited from [9]2022.03 | 55.3 | — | — | — | — | — | — | 81.4 | 61.1 | — | — | |
| InsDisArchitecture=Faster R-CNN R50-C4, Pre-train data=ImageNet, Epoch=200, Protocol=Fine-tuned2022.03 | 55.2 | — | — | — | — | — | — | 80.9 | 61.2 | — | — | |
| InstDisBackbone=ResNet-50-C42024.10 | 55.2 | — | — | — | — | — | — | 80.9 | 61.2 | — | — | |
| BYOLArchitecture=Faster R-CNN R50-C4, Pre-train data=COCO, Epoch=800, Protocol=Fine-tuned2022.03 | 53.8 | — | — | — | — | — | — | 79.9 | 59.1 | — | — | |
| SupervisedEpoch=902021.02 | 53.5 | — | — | — | — | — | — | 81.3 | 58.8 | — | — | |
| supervisedpretrain=supervised2021.10 | 53.5 | — | — | — | — | — | — | 81.3 | 58.8 | — | — | |
| SupervisedArchitecture=Faster R-CNN R50-C4, Pre-train data=ImageNet, Epoch=90, Protocol=Fine-tuned2022.03 | 53.5 | — | — | — | — | — | — | 81.3 | 58.8 | — | — | |
| SupervisedEpochs=90, Backbone=ResNet-50 C4, Architecture=Faster R-CNN2022.03 | 53.5 | — | — | — | — | — | — | 81.3 | 58.8 | — | — | |
| SupervisedBackbone=ResNet-50-C42024.10 | 53.5 | — | — | — | — | — | — | 81.3 | 58.8 | — | — | |
| Full Dataset#. Images=16,551, Backbone=ResNet-50, Detector=Faster R-CNN-C42025.02 | 52.16 | — | — | — | — | — | — | 80.63 | 56.92 | — | — | |
| BYOLEpoch=3002021.02 | 51.9 | — | — | — | — | — | — | 81 | 56.5 | — | — | |
| BYOLBackbone=ResNet-50-C4, Epochs=3002024.10 | 51.9 | — | — | — | — | — | — | 81 | 56.5 | — | — | |
| SimCLREpoch=2002021.02 | 51.5 | — | — | — | — | — | — | 79.4 | 55.6 | — | — | |
| Relative Loc.Epoch=2002021.02 | 50.6 | — | — | — | — | — | — | 76.9 | 55.2 | — | — | |
| SwAVEpoch=4002021.02 | 45.1 | — | — | — | — | — | — | 77.4 | 46.5 | — | — | |
| PCA#. Images=1,000, Backbone=ResNet-50, Detector=Faster R-CNN-C42025.02 | 35.99 | — | — | — | — | — | — | 65.89 | 34.86 | — | — | |
| Random initEpoch=None2021.02 | 33.8 | — | — | — | — | — | — | 60.2 | 33.1 | — | — | |
| Rand InitArchitecture=Faster R-CNN R50-C4, Pre-train data=None, Protocol=Fine-tuned2022.03 | 33.8 | — | — | — | — | — | — | 60.2 | 33.1 | — | — | |
| Random initBackbone=ResNet-50-C42024.10 | 33.8 | — | — | — | — | — | — | 60.2 | 33.1 | — | — | |
| VPS†#. Images=1,655, Backbone=ResNet-50, Detector=Faster R-CNN-C42025.02 | 30.05 | — | — | — | — | — | — | 60.77 | 25.84 | — | — | |
| AUM*#. Images=1,000, Backbone=ResNet-50, Detector=Faster R-CNN-C42025.02 | 22.31 | — | — | — | — | — | — | 46.36 | 18.23 | — | — | |
| Random#. Images=1,000, Backbone=ResNet-50, Detector=Faster R-CNN-C42025.02 | 21.17 | — | — | — | — | — | — | 44.97 | 17.26 | — | — |