Image Captioning on MS COCO (test) (Preference Scoring)
39.2Preference ScoreVisualGPT
Evaluation results
| Method | Links | |
|---|---|---|
| VisualGPTTraining data quantity=0.1%, Number of decoder layers=32021.02 | 39.2 | |
| VisualGPTTraining data quantity=0.5%, Number of decoder layers=32021.02 | 39.1 | |
| VisualGPTTraining data quantity=1%, Number of decoder layers=32021.02 | 37.4 | |
| M² TransformerTraining data quantity=0.1%, Number of decoder layers=32021.02 | 30.9 | |
| AoA TransformerTraining data quantity=1%, Number of decoder layers=32021.02 | 25 | |
| M² TransformerTraining data quantity=0.5%, Number of decoder layers=32021.02 | 22.8 | |
| AoA TransformerTraining data quantity=0.5%, Number of decoder layers=32021.02 | 20.9 | |
| M² TransformerTraining data quantity=1%, Number of decoder layers=32021.02 | 20.8 | |
| TransformerTraining data quantity=0.1%, Number of decoder layers=32021.02 | 18.4 | |
| TransformerTraining data quantity=0.5%, Number of decoder layers=32021.02 | 17.2 | |
| TransformerTraining data quantity=1%, Number of decoder layers=32021.02 | 16.8 | |
| AoA TransformerTraining data quantity=0.1%, Number of decoder layers=32021.02 | 11.5 |