TABLE 1.
Comparison of model performance on MS COCO dataset.
| Models | BLEU-1 | BLEU-4 | METEOR | ROUGE-L | CIDEr |
|---|---|---|---|---|---|
| Google NIC | 0.666 | 0.277 | 0.237 | - | 0.855 |
| Soft attention | 0.707 | 0.243 | 0.239 | - | - |
| g-LSTM | 0.670 | 0.264 | 0.227 | - | 0.813 |
| RIC | 0.734 | 0.299 | 0.254 | - | - |
| RHN | 0.723 | 0.306 | 0.252 | - | 0.989 |
| LSTM-A5 | 0.730 | 0.325 | 0.251 | 0.538 | 0.986 |
| This paper (basic model with no DRL and attention mechanism) | 0.716 | 0.289 | 0.244 | 0.456 | 0.893 |
| This paper (final model with DRL but no attention mechanism) | 0.746 | 0.339 | 0.284 | 0.583 | 0.991 |
| This paper (final model with DRL and attention mechanism) | 0.752 | 0.344 | 0.289 | 0.588 | 1.066 |