Vision-and-Language Navigation on Room-to-Room (R2R) Unseen (test)
8.13TLSeq2seq
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Seq2seq2019.03 | 8.13 | 7.85 | 20 | 18 | |
| RANDOM2019.03 | 9.93 | 9.77 | 13 | 12 | |
| Human2019.03 | 11.85 | 1.61 | 86 | 76 | |
| SPEAKER-FOLLOWERSearch Strategy=Greedy2019.03 | 14.82 | 6.62 | 35 | 28 | |
| SMNASearch Strategy=Greedy2019.03 | 18.04 | 5.67 | 48 | 35 | |
| + FASTSearch Strategy=Greedy, Trajectory length version=short2019.03 | 22.08 | 5.14 | 54 | 41 | |
| + FASTSearch Strategy=Beam, Trajectory length version=long2019.03 | 196.53 | 4.29 | 61 | 3 | |
| SMNASearch Strategy=Beam2019.03 | 373.09 | 4.48 | 61 | 2 | |
| SPEAKER-FOLLOWERSearch Strategy=Beam2019.03 | 1,257.3 | 4.87 | 53 | 1 |