Structured Vision-Language Concept Learning on Visual Genome (VG) 7-task sequence
86.88Final AccuracyOurs
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OursNprm=2.8%2022.11 | 86.88 | 93 | 1.12 | |
| Ours^{t-1}Nprm=2.8%2022.11 | 85.87 | 92.29 | 1.83 | |
| CFT-FHNprm=99.7%2022.11 | 81.3 | 85.51 | 5.88 | |
| BLIP2022.11 | 79.58 | 74.24 | — | |
| CFTNprm=100%2022.11 | 77.99 | 83.5 | 6.94 | |
| L2Nprm=100%2022.11 | 77.4 | 87.12 | 5.21 | |
| OursrandNprm=2.8%2022.11 | 77.2 | 88.62 | 3.7 | |
| LwFNprm=100%2022.11 | 77.01 | 85.24 | 5.8 | |
| EWCNprm=100%2022.11 | 72.78 | 82.61 | 7.23 | |
| DPNprm=≤ 5%2022.11 | 65.55 | 66.94 | 7.05 | |
| CFT-LPNprm=0.3%2022.11 | 65.13 | 72.7 | 11.73 | |
| L2PNprm=≤ 5%2022.11 | 60.88 | 69.23 | 17.07 |