Scene Graph Detection on VG150 (test)
19.8ng-mR@50MOTIFS-L2+cKD
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MOTIFS-L2+cKDBackbone=ResNeXt-101-FPN2021.07 | 19.8 | 14.2 | — | |
| MOTIFS-L2+uKDBackbone=ResNeXt-101-FPN2021.07 | 19.5 | 14 | — | |
| VCTree-L2+uKDBackbone=ResNeXt-101-FPN2021.07 | 19.1 | 13.8 | — | |
| VCTree-L2+cKDBackbone=ResNeXt-101-FPN2021.07 | 19 | 13.9 | — | |
| MOTIFS-DLFEBackbone=ResNeXt-101-FPN2021.07 | 18.1 | 11.7 | 23 | |
| SHA + GCL2022.03 | 17.9 | 14.2 | 20.9 | |
| VCTree-DLFEBackbone=ResNeXt-101-FPN2021.07 | 17.5 | 11.7 | 22.5 | |
| Motifs + GCL2022.03 | 16.8 | 12.9 | 19.3 | |
| GB-Net-βBackbone=VGG16, External knowledge bases=true2021.07 | 16.6 | 11.7 | — | |
| KERNBackbone=VGG162021.07 | 16 | 11.7 | — | |
| VCTree + GCL2022.03 | 15.2 | 11.9 | 17.5 | |
| VCTree-PCPLBackbone=ResNeXt-101-FPN2021.07 | 15.1 | 9.9 | 19.9 | |
| MOTIFS-PCPLBackbone=ResNeXt-101-FPN2021.07 | 14.8 | 9.8 | 19.6 | |
| MOTIFS-ReweightBackbone=ResNeXt-101-FPN2021.07 | 12.9 | 8 | 16.8 | |
| MOTIFSBackbone=ResNeXt-101-FPN2021.07 | 12.5 | 7.5 | 16.9 | |
| VCTreeBackbone=ResNeXt-101-FPN2021.07 | 12.5 | 7.5 | 16.7 | |
| VCTree-ReweightBackbone=ResNeXt-101-FPN2021.07 | 12.1 | 8 | 15.9 | |
| VCTree + DLFEBias Handling Strategy=Model-agnostic2022.03 | 11.8 | 8.6 | 13.8 | |
| Motifs + DLFEBias Handling Strategy=Model-agnostic2022.03 | 11.7 | 8.6 | 13.8 | |
| VCTree-TDEBackbone=ResNeXt-101-FPN2021.07 | 11.5 | 7.8 | 15.2 | |
| MOTIFS-TDEBackbone=ResNeXt-101-FPN2021.07 | 11.2 | 7.4 | 14.9 | |
| SG-CogTree2022.03 | 11.1 | 7.9 | 12.7 | |
| BGNN2022.03 | 10.7 | — | 12.6 | |
| MOTIFS-STLBackbone=ResNeXt-101-FPN2021.07 | 10.6 | 6.4 | 15 | |
| Motifs + CogTreeBias Handling Strategy=Model-agnostic2022.03 | 10.4 | 7.9 | 11.8 | |
| VCTree + CogTreeBias Handling Strategy=Model-agnostic2022.03 | 10.4 | 7.8 | 12.1 | |
| VCTree-STLBackbone=ResNeXt-101-FPN2021.07 | 10 | 6 | 14.1 | |
| PCPL+Object Detector=Faster R-CNN, Backbone=VGG-162022.03 | 9.5 | — | 11.7 | |
| VCTree + TDEBias Handling Strategy=Model-agnostic2022.03 | 9.3 | 6.9 | 11.1 | |
| Motifs + ReweightBias Handling Strategy=Model-agnostic2022.03 | 9.2 | 6.7 | 10.9 | |
| GPS-Net2022.03 | 8.7 | 6.9 | 9.8 | |
| VCTree + ReweightBias Handling Strategy=Model-agnostic2022.03 | 8.7 | 6.6 | 10.1 | |
| Motifs + TDEBias Handling Strategy=Model-agnostic2022.03 | 8.2 | 5.8 | 9.8 | |
| SHA2022.03 | 7.8 | 5.7 | 9.1 | |
| Motifs + EBMBias Handling Strategy=Model-agnostic2022.03 | 7.7 | 5.7 | 9.3 | |
| VCTree + EBMBias Handling Strategy=Model-agnostic2022.03 | 7.7 | 5.7 | 9.1 | |
| VCTree2022.03 | 7.4 | 5.4 | 8.7 | |
| VTransE+Object Detector=Faster R-CNN, Backbone=VGG-162022.03 | 6.8 | 5.1 | 8 | |
| Motifs2022.03 | 6.8 | 5 | 7.9 | |
| KERN2022.03 | 6.4 | — | 7.3 | |
| IMP+Object Detector=Faster R-CNN, Backbone=VGG-162022.03 | 3.8 | — | 4.8 |