Next step prediction on Crafts domain (test)
21.3BLEU-1BART+CP+M+R+CL
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| BART+CP+M+R+CLContextual Prefix=enabled, Selective Multimedia Encoder=enabled, Retrieval-Augmented Decoder=enabled, Diversity-Oriented Contrastive Learning=enabled2022.08 | 21.3 | 11.8 | 7.12 | 4.85 | 11.25 | 20.3 | -4.26 | 0.313 | |
| T52022.08 | 20.8 | 11.1 | 6.43 | 4.07 | 10.54 | 19.6 | -4.38 | 0.3 | |
| BART+CP+M+RContextual Prefix=enabled, Selective Multimedia Encoder=enabled, Retrieval-Augmented Decoder=enabled2022.08 | 20.7 | 11.5 | 6.93 | 4.66 | 11.02 | 20.5 | -4.25 | 0.309 | |
| BART+CP+MContextual Prefix=enabled, Selective Multimedia Encoder=enabled2022.08 | 20.5 | 11.1 | 6.61 | 4.4 | 10.79 | 20.1 | -4.28 | 0.305 | |
| BART+CPContextual Prefix=enabled2022.08 | 20.1 | 11.1 | 6.48 | 4.24 | 10.61 | 20.1 | -4.29 | 0.303 | |
| BART2022.08 | 19.7 | 10.8 | 6.22 | 4.11 | 10.44 | 20 | -4.29 | 0.299 | |
| Retrieval BART2022.08 | 18.7 | 9.78 | 5.52 | 3.52 | 9.89 | 18.2 | -4.38 | 0.285 | |
| CLIP-BARTBackbone=ViT-B/32 CLIP encoder2022.08 | 17.9 | 9.13 | 5.21 | 3.4 | 9.37 | 16.4 | -4.56 | 0.245 | |
| GPT-22022.08 | 15.5 | 5.4 | 1.98 | 0.93 | 7.63 | 14 | -4.67 | 0.218 | |
| GPT2-SIFTraining Mode=fine-tuned2022.08 | 14.8 | 6.7 | 3.05 | 1.58 | 7.74 | 13.2 | -4.69 | 0.234 | |
| Naive Retrieval2022.08 | 13.5 | 5.26 | 2.38 | 1.28 | 6.81 | 12.3 | -4.83 | 0.163 |