Cross-modal Matching on NEWSTT (test)
81.5F1 ScoreCFT-CLIP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CFT-CLIPPretraining Dataset=BBC, Reference Text=Summary, Targeting=Person-labeled entity tokens2024.02 | 81.5 | 0.491 | |
| CLIPAdaptPretraining Dataset=BBC, Reference Text=Summary, Objective=CLIP objective2024.02 | 76.7 | 0.459 | |
| CLIPModel Type=Pretrained, Parameters=427M2024.02 | 76.3 | 0.409 | |
| BLIPModel Type=Pretrained2024.02 | 73.7 | 0.408 | |
| BLIP-2Model Type=Pretrained, Parameters=1.17B2024.02 | 70.7 | 0.415 | |
| BLIP-2 + SBERTPipeline=BLIP-2 captioning + SentenceBERT similarity2024.02 | 69.4 | 0.341 |