We follow the same evaluation protocols and metrics with Scene-Graph-Benchmark.pytorch(Visual Genome dataset) and Graphical Contrastive Losses for Scene Graph Parsing (Openimage V4/V6)
Here we list the SGGen results can be produced by our codebase with X-101-FPN backbone. We reimplement those methods according the official implementation released by author.
∗ denotes the LVIS resampling is applied for this model.
| Model(SGGen) | mR@50 | mR@100 | R@50 | R@100 | head | body | tail |
|---|---|---|---|---|---|---|---|
| RelDN | 6.0 | 7.3 | 31.4 | 35.9 | 34.1 | 6.6 | 1.1 |
| Motifs | 5.5 | 6.8 | 32.1 | 36.9 | 36.1 | 7.0 | 0.0 |
| Motifs∗ | 7.7 | 9.4 | 31.7 | 35.8 | 34.2 | 8.6 | 2.1 |
| VCTree | 10.9 | 13.5 | 29.8 | 34.6 | - | - | - |
| G-RCNN | 5.8 | 6.7 | 29.78 | 32.8 | 28.6 | 6.5 | 0.1 |
| MSDN | 6.1 | 7.2 | 31.9 | 36.6 | 35.1 | 5.5 | 0.0 |
| Unbiased | 9.3 | 11.1 | 19.4 | 23.2 | 24.5 | 13.9 | 0.1 |
| GPS-Net | 6.79 | 8.6 | 31.1 | 35.9 | 34.5 | 7.0 | 1.0 |
| GPS-Net* | 7.4 | 9.5 | 27.8 | 32.1 | 30.4 | 8.5 | 3.8 |
| BGNN | 10.9 | 13.55 | 29.8 | 34.6 | 33.4 | 13.4 | 6.4 |
| Model(SGGen) | mR@50 | R@50 | wmAP_rel | wmAP_phr | score_wtd |
|---|---|---|---|---|---|
| RelDN | 33.98 | 73.08 | 32.16 | 33.39 | 40.84 |
| RelDN* | 37.20 | 75.34 | 33.21 | 34.31 | 41.97 |
| VCTree | 33.91 | 74.08 | 34.16 | 33.11 | 40.21 |
| G-RCNN | 34.04 | 74.51 | 33.15 | 34.21 | 41.84 |
| Motifs | 32.68 | 71.63 | 29.91 | 31.59 | 38.93 |
| Unbiased | 35.47 | 69.30 | 30.74 | 32.80 | 39.27 |
| GPS-Net | 35.26 | 74.81 | 32.85 | 33.98 | 41.69 |
| GPS-Net* | 38.93 | 74.74 | 32.77 | 33.87 | 41.60 |
| BGNN | 41.71 | 74.96 | 33.83 | 34.87 | 42.47 |