Explanation Generation On Whoops
評価指標
Human (%)
評価結果
このベンチマークにおける各モデルのパフォーマンス結果
モデル名 | Human (%) | Paper Title | Repository |
|---|---|---|---|
| Predicted Caption -> GPT3 | 33 | Breaking Common Sense: WHOOPS! A Vision-and-Language Benchmark of Synthetic and Compositional Images | - |
| VLIS (LLaVA) | - | VLIS: Unimodal Language Models Guide Multimodal Language Generation | - |
| BLIP2 FlanT5-XL (Fine-tuned) | 15 | Breaking Common Sense: WHOOPS! A Vision-and-Language Benchmark of Synthetic and Compositional Images | - |
| BLIP2 FlanT5-XXL (Fine-tuned) | 27 | Breaking Common Sense: WHOOPS! A Vision-and-Language Benchmark of Synthetic and Compositional Images | - |
| VLIS (Lynx) | - | VLIS: Unimodal Language Models Guide Multimodal Language Generation | - |
| Ground-truth Caption -> GPT3 (Oracle) | 68 | Breaking Common Sense: WHOOPS! A Vision-and-Language Benchmark of Synthetic and Compositional Images | - |
| BLIP2 FlanT5-XXL (Zero-shot) | 0 | Breaking Common Sense: WHOOPS! A Vision-and-Language Benchmark of Synthetic and Compositional Images | - |
0 of 7 row(s) selected.