Scene Graph Generation on Visual Genome VG150 (test)
41.3RecallOvSGTR
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| OvSGTRMethod Type=VLM-based2026.03 | 41.3 | — | — | — | — | — | — | — | — | 8.8 | 25.1 | — | |
| VS3Method Type=VLM-based2026.03 | 40.9 | — | — | — | — | — | — | — | — | 7.8 | 24.4 | — | |
| DSGGMethod Type=non-VLM2026.03 | 38.5 | — | — | — | — | — | — | — | — | 17.3 | 27.9 | 3.9 | |
| SGG-R3Params=3B, Base Model=Qwen2.5-VL, Training Setting=SFT + RL2026.03 | 36 | — | — | — | — | — | — | — | — | 14.8 | 25.4 | 6.1 | |
| PE-NetMethod Type=non-VLM2026.03 | 35.2 | — | — | — | — | — | — | — | — | 14.5 | 24.9 | 3.6 | |
| EGTRMethod Type=non-VLM2026.03 | 34.3 | — | — | — | — | — | — | — | — | 10.1 | 22.2 | — | |
| GPS-NetMethod Type=non-VLM2026.03 | 31.7 | — | — | — | — | — | — | — | — | 9.8 | 20.8 | — | |
| VCTreeMethod Type=non-VLM2026.03 | 31.3 | — | — | — | — | — | — | — | — | 8 | 19.7 | 0.7 | |
| RelTRMethod Type=non-VLM2026.03 | 30.7 | — | — | — | — | — | — | — | — | 12.6 | 21.7 | 2.4 | |
| MOTIFSMethod Type=non-VLM2026.03 | 30.3 | — | — | — | — | — | — | — | — | 6.6 | 18.5 | 0.2 | |
| SGTR+Method Type=non-VLM2026.03 | 30.1 | — | — | — | — | — | — | — | — | 17 | 23.6 | 5.8 | |
| Pair-NetMethod Type=non-VLM2026.03 | 29.3 | — | — | — | — | — | — | — | — | 15.4 | 22.4 | — | |
| HiLoMethod Type=non-VLM2026.03 | 29.2 | — | — | — | — | — | — | — | — | 17.6 | 23.4 | — | |
| SGTRMethod Type=non-VLM2026.03 | 28.4 | — | — | — | — | — | — | — | — | 15.2 | 21.8 | 5.8 | |
| R1-SGGMethod Type=VLM-based, Params=7B, Base Model=Qwen2-VL2026.03 | 27.6 | — | — | — | — | — | — | — | — | 10.9 | 19.3 | 4.4 | |
| PGSGMethod Type=VLM-based2026.03 | 23.6 | — | — | — | — | — | — | — | — | 12.7 | 18.2 | 7.6 | |
| SGG-R3Params=3B, Base Model=Qwen2.5-VL, Training Setting=SFT2026.03 | 17.5 | — | — | — | — | — | — | — | — | 8.2 | 12.9 | 2.4 | |
| OwSGGMethod Type=VLM-based, Params=72B, Base Model=Qwen2-VL2026.03 | 3.2 | — | — | — | — | — | — | — | — | 3.4 | 3.3 | 2 | |
| AS-NetBackbone=ResNet-1012024.03 | — | 18.7 | 21.1 | 6.1 | 7.2 | 12.4 | 14.2 | 9.2 | 10.7 | — | — | — | |
| BGNNBackbone=ResNet-1012024.03 | — | 28.2 | 33.8 | 8.6 | 10.3 | 18.4 | 22.1 | 13.2 | 15.8 | — | — | — | |
| DT2-ACBSBackbone=X101-FPN2024.03 | — | 22 | 24.4 | 15 | 16.3 | 18.5 | 20.4 | 17.8 | 19.5 | — | — | — | |
| HOTRBackbone=ResNet-101, Reproduced result (*)=true2024.03 | — | 22.4 | 27.1 | 6.9 | 9.7 | 14.7 | 18.4 | 10.6 | 14.3 | — | — | — | |
| HOTR + SpeaQBackbone=ResNet-101, Reproduced result (*)=true2024.03 | — | 24.7 | 29.1 | 9.6 | 12.7 | 17.2 | 20.9 | 13.8 | 17.7 | — | — | — | |
| ISGBackbone=ResNet-101, Without loss re-weighting (†)=true2024.03 | — | 29.5 | 32.1 | 7.4 | 8.4 | 18.5 | 20.3 | 11.8 | 13.3 | — | — | — | |
| ISGBackbone=ResNet-101, Reproduced result (*)=true2024.03 | — | 27.2 | 30.1 | 15 | 16.6 | 21.1 | 23.4 | 19.3 | 21.4 | — | — | — | |
| ISG + SpeaQBackbone=ResNet-101, Without loss re-weighting (†)=true2024.03 | — | 32.9 | 36 | 11.8 | 14.1 | 22.4 | 25.1 | 17.4 | 20.3 | — | — | — | |
| ISG + SpeaQBackbone=ResNet-101, Reproduced result (*)=true2024.03 | — | 32.1 | 35.5 | 15.1 | 17.6 | 23.6 | 26.6 | 20.5 | 23.5 | — | — | — | |
| MotifsBackbone=X101-FPN2024.03 | — | 32.1 | 36.9 | 5.5 | 6.8 | 18.8 | 21.9 | 9.4 | 11.5 | — | — | — | |
| RelDNBackbone=ResNet-1012024.03 | — | 30.3 | 34.8 | 4.4 | 5.4 | 17.4 | 20.1 | 7.7 | 9.3 | — | — | — | |
| SGTRBackbone=ResNet-1012024.03 | — | 25.1 | 26.6 | 12 | 14.6 | 18.6 | 20.6 | 16.2 | 18.9 | — | — | — | |
| VCTreeBackbone=X101-FPN2024.03 | — | 31.8 | 36.1 | 6.6 | 7.7 | 19.2 | 21.9 | 10.9 | 12.7 | — | — | — | |
| VCTree-BPLSABackbone=X101-FPN2024.03 | — | 21.7 | 25.5 | 13.5 | 15.7 | 17.6 | 20.6 | 16.6 | 19.4 | — | — | — | |
| VCTree-EBMBackbone=X101-FPN2024.03 | — | 20.5 | 24.7 | 9.7 | 11.6 | 15.1 | 18.2 | 13.2 | 15.8 | — | — | — | |
| VCTree-TDEBackbone=X101-FPN2024.03 | — | 19.4 | 23.2 | 9.3 | 11.1 | 14.4 | 17.2 | 12.6 | 15 | — | — | — |