Scene Graph Generation on Visual Genome (test)
30.9R@50KERN
Evaluation Results
| Method | Links | ||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| KERNConstraint=false2019.03 | 30.9 | — | 35.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SMNConstraint=false2019.03 | 30.5 | — | 35.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VCTREE-HLdetector=Faster-RCNN, backbone=VGG2018.12 | 27.9 | 22 | 31.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VCTREE-SLdetector=Faster-RCNN, backbone=VGG2018.12 | 27.7 | 21.7 | 31.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Overlapdetector=Faster-RCNN, backbone=VGG2018.12 | 27.3 | 21.4 | 30.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Multi-Branchdetector=Faster-RCNN, backbone=VGG2018.12 | 27.3 | 21.5 | 30.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MOTIFSdetector=Faster-RCNN, backbone=VGG2018.12 | 27.2 | 21.4 | 30.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SMNConstraint=true2019.03 | 27.2 | — | 30.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Chaindetector=Faster-RCNN, backbone=VGG2018.12 | 27.1 | 21.2 | 30.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| KERNConstraint=true2019.03 | 27.1 | — | 29.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FREQdetector=Faster-RCNN, backbone=VGG2018.12 | 26.2 | 20.1 | 30.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FREQConstraint=false2019.03 | 25.3 | — | 30.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FREQConstraint=true2019.03 | 23.5 | — | 27.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMP+Constraint=false2019.03 | 22 | — | 27.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMPdetector=Faster-RCNN, backbone=VGG2018.12 | 20.7 | 14.6 | 24.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMP+Constraint=true2019.03 | 20.7 | — | 24.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| KERNConstraint=false2019.03 | 11.7 | — | 16 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Graph-RCNN2018.12 | 11.4 | — | 13.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AEConstraint=false2019.03 | 9.7 | — | 11.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SMNConstraint=false2019.03 | 9.3 | — | 12.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AsscEmbed2018.12 | 8.1 | 6.5 | 8.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| KERNConstraint=true2019.03 | 6.4 | — | 7.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FREQConstraint=false2019.03 | 5.9 | — | 8.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMP+Constraint=false2019.03 | 5.4 | — | 8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SMNConstraint=true2019.03 | 5.3 | — | 6.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TFR2018.12 | 4.8 | 3.4 | 6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FREQConstraint=true2019.03 | 4.3 | — | 5.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMP+Constraint=true2019.03 | 3.8 | — | 4.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMPConstraint=true2019.03 | 3.4 | — | 4.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AEConstraint=false2019.03 | 1.6 | — | 2.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMPConstraint=true2019.03 | 0.6 | — | 0.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MOTIFSLocation-Free=false2023.03 | 0.3808 | 0.2902 | 0.4364 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransformerLocation-Free=false2023.03 | 0.3781 | 0.2879 | 0.4369 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPS-NetStage=two-stage, Backbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN2024.04 | 0.359 | 0.311 | — | 6.7 | 8.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VCTreeLocation-Free=false2023.03 | 0.3559 | 0.2706 | 0.4121 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HQSG*†Backbone=ResNet-101, loss-reweighting strategy=true, Qtrain=6002026.06 | 0.341 | — | 0.383 | — | 16 | 20.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | 21.8 | 26.7 | — | — | — | |
| DRM w/o DKTBackbone=X101-FPN, Unbiased=false2026.07 | 0.34 | — | 0.389 | — | 9 | 11.2 | — | — | 14.2 | 17.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SSR-CNNBackbone=X101-FPN, #params(M)=274.3, FPS=3.1, Unbiased=false2026.07 | 0.335 | — | 0.384 | — | 8.6 | 10.3 | — | — | 13.7 | 16.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Dual-SGGBackbone=ResNet, #params(M)=84.7, FPS=14.0, Unbiased=false2026.07 | 0.335 | — | 0.385 | — | 10.3 | 12.3 | — | — | 15.8 | 18.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MDMBackbone=ResNet-101, Qeval=300, λ=0.0, Tbox=4, Tlabel=42026.06 | 0.329 | — | 0.373 | — | 13.5 | 15.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | 19.1 | 22.3 | 32.2 | 23.6 | 9.8 | |
| MotifsBackbone=ResNeXt101-FPN, # params (M)=369.9, architecture=two-stage, debiasing_strategy=false2026.01 | 0.328 | 0.255 | 0.372 | 5 | 6.8 | 7.9 | — | 8.4 | 11.3 | 13 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MDMBackbone=ResNet-101, Qeval=300, λ=0.1, Tbox=4, Tlabel=42026.06 | 0.328 | — | 0.373 | — | 15.3 | 17.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | 20.9 | 24.2 | 33.8 | 23.8 | 7.2 | |
| SSR-CNNStage=one-stage, # params (M)=274.3, FPS=4.0, AP50 NMS=0.3 (concatenated subjects/objects)2024.04 | 0.327 | 0.258 | 0.369 | 6.1 | 8.4 | 10 | 23.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PE-NetBackbone=X101-FPN, Unbiased=false2026.07 | 0.324 | — | 0.369 | — | 8.9 | 11 | — | — | 14 | 16.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MDMBackbone=ResNet-101, Qeval=300, λ=0.2, Tbox=4, Tlabel=42026.06 | 0.323 | — | 0.367 | — | 16.7 | 19.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | 22 | 25.6 | 32.9 | 25.8 | 9.7 | |
| MotifsStage=two-stage, Backbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN, # params (M)=369.9, FPS=1.92024.04 | 0.321 | 0.251 | 0.369 | 4.1 | 5.5 | 6.8 | 28.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpeaQBackbone=ResNet101, # params (M)=93.4, architecture=one-stage2026.01 | 0.321 | 0.251 | 0.355 | 10.1 | 15.1 | 17.6 | — | 14.4 | 20.5 | 23.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MOTIFBackbone=X101-FPN2026.06 | 0.321 | — | 0.369 | — | 5.5 | 6.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | 9.4 | 11.5 | — | — | — | |
| SpeaQ* ISGBackbone=ResNet-101, loss-reweighting strategy=true2026.06 | 0.321 | — | 0.355 | — | 15.1 | 17.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | 20.5 | 23.5 | — | — | — | |
| MotifsBackbone=X101-FPN, #params(M)=369.9, FPS=1.9, Unbiased=false2026.07 | 0.321 | — | 0.369 | — | 5.5 | 6.8 | — | — | 9.4 | 11.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VCTreeStage=two-stage, Backbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN, # params (M)=361.5, FPS=0.82024.04 | 0.318 | 0.248 | 0.361 | 4.9 | 6.6 | 7.7 | 28.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VCTreeBackbone=X101-FPN2026.06 | 0.318 | — | 0.361 | — | 6.6 | 7.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | 10.9 | 12.7 | — | — | — | |
| VCTreeBackbone=X101-FPN, #params(M)=361.5, FPS=0.8, Unbiased=false2026.07 | 0.318 | — | 0.361 | — | 6.6 | 7.7 | — | — | 10.9 | 12.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MDMBackbone=ResNet-101, Qeval=300, λ=0.3, Tbox=4, Tlabel=42026.06 | 0.316 | — | 0.361 | — | 18.3 | 21.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | 23.2 | 26.9 | 31.8 | 27.2 | 12.6 | |
| MDMBackbone=ResNet-101, Qeval=900, λ=0.4, Tbox=4, Tlabel=42026.06 | 0.315 | — | 0.361 | — | 20.2 | 23.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | 24.6 | 28.5 | 31.2 | 29.2 | 15.6 | |
| VCTree (EBM)Stage=two-stage, Backbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN, # params (M)=372.52024.04 | 0.314 | 0.242 | 0.359 | 5.7 | 7.7 | 9.1 | 28.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RelDNBackbone=X101-FPN2026.06 | 0.314 | — | 0.359 | — | 6 | 7.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | 10.1 | 12.1 | — | — | — | |
| VTransEStage=two-stage, Backbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN, # params (M)=312.3, FPS=3.52024.04 | 0.313 | 0.245 | 0.355 | 5.1 | 6.8 | 8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MDMBackbone=ResNet-101, Qeval=600, λ=0.4, Tbox=4, Tlabel=42026.06 | 0.312 | — | 0.357 | — | 20.1 | 23.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | 24.4 | 28.3 | 31 | 29 | 15.8 | |
| BGNNStage Type=Two-Stage, Extra Features=Yes, Backbone=X-101FPN, Number of Parameters (M)=341.92023.06 | 0.31 | 0.233 | 0.346 | 7.5 | 10.7 | 13.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BGNNStage=two-stage, Backbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN, # params (M)=341.9, FPS=1.72024.04 | 0.31 | 0.233 | 0.358 | 7.5 | 10.7 | 12.6 | 29 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BGNNBackbone=ResNeXt101-FPN, # params (M)=341.9, architecture=two-stage2026.01 | 0.31 | 0.233 | 0.358 | 7.5 | 10.7 | 12.6 | — | 11.3 | 15.9 | 18.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BGNNBackbone=X101-FPN2026.06 | 0.31 | — | 0.358 | — | 10.7 | 12.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | 15.9 | 18.6 | 34 | 12.9 | 6 | |
| BGNNBackbone=X101-FPN, #params(M)=341.9, FPS=1.7, Unbiased=true2026.07 | 0.31 | — | 0.358 | — | 10.7 | 12.6 | — | — | 15.9 | 18.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RelTRLocation-Free=false2023.03 | 0.3099 | 0.2586 | 0.3331 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMPLocation-Free=false2023.03 | 0.3078 | 0.2166 | 0.3707 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PE-NetBackbone=X101-FPN2026.06 | 0.307 | — | 0.352 | — | 12.4 | 14.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | 17.7 | 20.5 | — | — | — | |
| SGTRLocation-Free=false2023.03 | 0.3038 | 0.2362 | 0.3485 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RelDNBackbone=ResNet-1012026.06 | 0.303 | — | 0.348 | — | 4.4 | 5.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | 7.7 | 9.3 | 31.3 | 2.3 | 0 | |
| EGTRStage=one-stage, # params (M)=42.5, FPS=14.72024.04 | 0.302 | 0.235 | 0.343 | 5.5 | 7.9 | 10.1 | 30.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MDMBackbone=ResNet-101, Qeval=300, λ=0.4, Tbox=4, Tlabel=-2026.06 | 0.302 | — | 0.343 | — | 18.6 | 21.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | 23 | 26.7 | 29.3 | 27.2 | 14.3 | |
| MDMBackbone=ResNet-101, Qeval=300, λ=0.4, Tbox=4, Tlabel=12026.06 | 0.302 | — | 0.345 | — | 19.3 | 22.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | 23.5 | 27.4 | 30.1 | 28.2 | 15 | |
| MDMBackbone=ResNet-101, Qeval=300, λ=0.4, Tbox=4, Tlabel=42026.06 | 0.302 | — | 0.346 | — | 19.7 | 22.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | 23.8 | 27.6 | 30.2 | 28.1 | 15.6 | |
| EGTRBackbone=ResNet, #params(M)=42.5, FPS=14.7, Unbiased=false2026.07 | 0.302 | — | 0.343 | — | 7.9 | 10.1 | — | — | 12.5 | 15.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Dual-SGG+LABackbone=ResNet, #params(M)=84.7, FPS=14.0, Unbiased=true, Logit Adjustment=true, tau=0.22026.07 | 0.302 | — | 0.352 | — | 18.7 | 22.2 | — | — | 23.1 | 27.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VRD2018.12 | 0.3 | — | 0.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VRDConstraint=true2019.03 | 0.3 | — | 0.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Iterative SGGStage=one-stage, # params (M)=93.5, FPS=6.0, AP50 NMS=0.3 (concatenated subjects/objects)2024.04 | 0.297 | — | 0.321 | — | 8 | 8.8 | 27.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ISGBackbone=ResNet, #params(M)=93.5, FPS=6.0, Unbiased=false2026.07 | 0.297 | — | 0.321 | — | 8 | 8.8 | — | — | 12.6 | 13.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MDMBackbone=ResNet-101, Qeval=300, λ=0.4, Tbox=1, Tlabel=-2026.06 | 0.292 | — | 0.328 | — | 18.3 | 21 | — | — | — | — | — | — | — | — | — | — | — | — | — | 22.5 | 25.6 | 28.4 | 26.1 | 13.8 | |
| Mg-RMPNBackbone=ResNet50, architecture=one-stage2026.01 | 0.291 | 0.225 | 0.335 | 10.3 | 14.4 | 17.3 | — | 14.1 | 19.3 | 22.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Mg-RMPNBackbone=ResNet, Unbiased=true2026.07 | 0.291 | — | 0.335 | — | 14.4 | 17.3 | — | — | 19.3 | 22.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPS-NetStage Type=Two-Stage, Extra Features=Yes, Backbone=VGG162023.06 | 0.289 | 0.223 | 0.332 | 6.9 | 8.7 | 9.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Salience-SGGBackbone=ResNet50, # params (M)=77.7, architecture=one-stage2026.01 | 0.288 | 0.219 | 0.334 | 12.8 | 18 | 21.6 | — | 16.2 | 22.1 | 26.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Salience-SGGBackbone=ResNet, #params(M)=77.7, FPS=13.3, Unbiased=true2026.07 | 0.288 | — | 0.334 | — | 18 | 21.6 | — | — | 22.1 | 26.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Hydra-SGGBackbone=ResNet50, # params (M)=67.6, architecture=one-stage2026.01 | 0.286 | 0.219 | 0.334 | 10.3 | 15.9 | 19.4 | — | 14 | 20.5 | 24.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VETOBackbone=X101-FPN2026.06 | 0.286 | — | 0.34 | — | 10.6 | 13.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | 15.5 | 19.6 | — | — | — | |
| Hydra-SGGBackbone=ResNet-502026.06 | 0.286 | — | 0.334 | — | 15.9 | 19.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | 20.5 | 24.7 | — | — | — | |
| RelationformerStage Type=One-Stage, Extra Features=No, Backbone=ResNet-50, Number of Parameters (M)=92.92023.06 | 0.284 | 0.222 | 0.313 | 4.6 | 9.3 | 10.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RelationformerStage=one-stage, # params (M)=92.9, FPS=8.52024.04 | 0.284 | 0.222 | 0.313 | 4.6 | 9.3 | 10.7 | 26.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RelationformerBackbone=ResNet, #params(M)=92.9, FPS=8.5, Unbiased=false2026.07 | 0.284 | — | 0.313 | — | 9.3 | 10.7 | — | — | 14 | 15.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Hydra-SGGBackbone=ResNet, #params(M)=67.6, FPS=5.3, Unbiased=true2026.07 | 0.284 | — | 0.331 | — | 16 | 19.7 | — | — | 20.5 | 24.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TraCQStage Type=One-Stage, Extra Features=No, Backbone=ResNet-50, Number of Parameters (M)=51.22023.06 | 0.283 | 0.197 | 0.357 | 12 | 13.8 | 14.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EGTRStage=one-stage, # params (M)=42.5, FPS=14.7, logit adjustment (LA)=true, threshold (T)=0.32024.04 | 0.282 | 0.224 | 0.317 | 8.8 | 14 | 18.3 | 30.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EGTRBackbone=ResNet50, # params (M)=42.5, architecture=one-stage2026.01 | 0.282 | 0.224 | 0.317 | 8.8 | 14 | 18.3 | — | 12.6 | 18.7 | 23.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BGNNBackbone=ResNet-1012026.06 | 0.282 | — | 0.338 | — | 8.6 | 10.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | 13.2 | 15.8 | 29.1 | 12.6 | 2.2 | |
| MDMBackbone=ResNet-101, Qeval=100, λ=0.4, Tbox=4, Tlabel=42026.06 | 0.282 | — | 0.32 | — | 18.8 | 21.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | 22.6 | 25.6 | 27.9 | 26.5 | 14.3 | |
| MDMBackbone=ResNet-101, Qeval=300, λ=0.5, Tbox=4, Tlabel=42026.06 | 0.277 | — | 0.317 | — | 20.7 | 24 | — | — | — | — | — | — | — | — | — | — | — | — | — | 23.7 | 27.3 | 27.2 | 28.7 | 18.5 | |
| RelTRStage=one-stage, # params (M)=63.7, FPS=13.42024.04 | 0.275 | 0.212 | — | 6.8 | 10.8 | — | 26.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RelTRBackbone=ResNet-502026.06 | 0.275 | — | 0.307 | — | 10.8 | 12.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | 15.5 | 17.6 | — | — | — |