Goal-Conditioned Visual Navigation Instruction Generation on R2R-Goal (val (Seen))
36BLEU-4Anole-7B + Interleaved
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Anole-7B + InterleavedReasoning Strategy=Interleaved2025.08 | 36 | 22 | 21 | 27 | |
| Anole-7B + One-passReasoning Strategy=One-pass2025.08 | 34 | 20 | 18 | 22 | |
| C-Instructor2025.08 | 21 | 19 | 14 | 23 | |
| Speaker-Follower2025.08 | 10 | 8 | 8 | 12 | |
| Claude 4 Opus2025.08 | 10 | 15 | 17 | 14 | |
| Anole-7B + CoTReasoning Strategy=fine-tuned CoT2025.08 | 10 | 14 | 15 | 17 | |
| Gemini 3.02025.08 | 9 | 13 | 16 | 13 | |
| GPT-4oReasoning Strategy=Direct2025.08 | 8 | 16 | 19 | 18 | |
| GPT-4o + CoTReasoning Strategy=zero-shot CoT2025.08 | 8 | 17 | 17 | 21 | |
| Gemini 2.02025.08 | 8 | 11 | 13 | 12 | |
| Claude 3 Opus2025.08 | 7 | 12 | 12 | 12 | |
| Anole-7B (Direct)Reasoning Strategy=Direct2025.08 | 6 | 10 | 10 | 14 | |
| LANA2025.08 | 5 | 5 | 11 | 10 |