Correlation with human judgment on Composite 1 (test)
57.3Kendall Tau-cRefPAC-S
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RefPAC-SReference-based=true2023.03 | 57.3 | 53 | |
| PAC-SReference-free=true2023.03 | 55.7 | 51.5 | |
| RefCLIP-SReference-based=true2023.03 | 55.4 | 51.2 | |
| CLIP-SReference-free=true2023.03 | 53.8 | 49.8 | |
| ViLBERTScoreReference-based=true2023.03 | 52.4 | — | |
| FAIErReference-based=true2023.03 | 51.4 | — | |
| TIGErReference-based=true2023.03 | 45.4 | — | |
| BERT-S++Reference-based=true2023.03 | 44.9 | — | |
| SPICEReference-based=true2023.03 | 40.3 | 38.8 | |
| METEORReference-based=true2023.03 | 38.9 | 36 | |
| CIDErReference-based=true2023.03 | 37.7 | 34.9 | |
| ROUGEReference-based=true2023.03 | 32.4 | 30 | |
| BLEU-1Reference-based=true2023.03 | 31.3 | 29 | |
| BLEU-4Reference-based=true2023.03 | 30.6 | 28.3 | |
| BERT-SReference-based=true2023.03 | 30.1 | — |