Scene Graph Generation on VG150 (test)
90.35Cov (Objects)BGNN + PC-SGG
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| BGNN + PC-SGGBackbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN, MLLM=BLIP-2 (FLAN-T5-XL) 8-bit, Evaluation Protocol=SGDET, alpha_o=0.05, alpha_r=0.12025.03 | 90.35 | 4.8 | 4.48 | 89.68 | 5.2 | 16.23 | 80.45 | |
| SQUAT + PC-SGGBackbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN, MLLM=BLIP-2 (FLAN-T5-XL) 8-bit, Evaluation Protocol=SGDET, alpha_o=0.05, alpha_r=0.12025.03 | 90.26 | 4.9 | 4.48 | 90.25 | 4.6 | 14.48 | 80.25 | |
| VCTREE + PC-SGGBackbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN, MLLM=BLIP-2 (FLAN-T5-XL) 8-bit, Evaluation Protocol=SGDET, alpha_o=0.05, alpha_r=0.12025.03 | 89.38 | 5.7 | 4.23 | 88.61 | 5.9 | 16.41 | 80.06 | |
| MOTIFS + PC-SGGBackbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN, MLLM=BLIP-2 (FLAN-T5-XL) 8-bit, Evaluation Protocol=SGDET, alpha_o=0.05, alpha_r=0.12025.03 | 88.94 | 5.8 | 4.87 | 84.11 | 6.2 | 16.09 | 74.97 | |
| MOTIFS-D + PC-SGGBackbone=ResNeXt-101-FPN, Object Detector=Faster R-CNN, MLLM=BLIP-2 (FLAN-T5-XL) 8-bit, Evaluation Protocol=SGDET, alpha_o=0.05, alpha_r=0.12025.03 | 88.94 | 5.8 | 4.87 | 86.67 | 5.9 | 16.81 | 76.67 |