Image Captioning on MS COCO random subset of 100 images (test)
40.7BLEU-4ClipCap
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ClipCapEvaluation protocol=finetuned, Training data supervision=full training set with image-text pairs2022.04 | 40.7 | 30.4 | 152.4 | 25.2 | 60.9 | |
| SMs 3-shotEvaluation protocol=few-shot (3-shot), Number of generated candidate captions (n)=202022.04 | 18.3 | 18.8 | 76.3 | 14.8 | 43.7 | |
| MAGICEvaluation protocol=finetuned on unpaired training set, Training data supervision=zero-shot on image-text pairs2022.04 | 11.4 | 16.4 | 56.2 | 11.3 | 39 | |
| SMs 0-shotEvaluation protocol=zero-shot, Number of generated candidate captions (n)=202022.04 | 6.9 | 15 | 44.5 | 10.1 | 34.1 | |
| ZeroCapEvaluation protocol=zero-shot2022.04 | 0 | 8.8 | 18 | 5.6 | 18.3 |