Region Captioning on Visual Genome (val)
17.6METEORGPT4RoI-13B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT4RoI-13BInput=Ground truth bounding boxes, fine-tuned_on_VG=true2023.07 | 17.6 | 11.7 | 35.2 | 146.8 | |
| GPT4RoI-7BInput=Ground truth bounding boxes, fine-tuned_on_VG=true2023.07 | 17.4 | 11.5 | 35 | 145.2 | |
| GRITInput=Ground truth bounding boxes2023.07 | 17.1 | — | — | 142 | |
| GPT4RoI-7BInput=Ground truth bounding boxes, fine-tuned_on_VG=false2023.07 | 16.5 | 10.5 | 33.4 | 134.5 | |
| Shikra-7BInput=Ground truth bounding boxes, concurrent_work=true2023.07 | 15.2 | 8.9 | 30.3 | 115.8 |