Image Captioning 벤치마크
Image Captioning on nocaps val
CIDEr
- 2021-02-04 — VL-T5: CIDEr 4.4
- 2022-06-13 — MetaLM: CIDEr 58.7
- 2023-03-04 — Prismer: CIDEr 107.9
| Rank | Model | CIDEr | SPICE | Paper | Code | Year |
|---|---|---|---|---|---|---|
| 1 | Prismer | 107.9 | 14.8 | Prismer: A Vision-Language Model with Multi-Task Experts | nvlabs/prismer · KastanDay/video-pretrained-transformer | 2023 |
| 2 | MetaLM | 58.7 | 8.6 | Language Models are General-Purpose Interfaces | microsoft/unilm | 2022 |
| 3 | VL-T5 | 4.4 | 5.3 | Unifying Vision-and-Language Tasks via Text Generation | j-min/VL-T5 · mitvis/vistext | 2021 |