Research map / Vision-language and multimodal / Robot action and navigation
Improving Vision-and-Language Navigation with Image-Text Pairs from the Web
Arjun Majumdar, Ayush Shrivastava, Stefan Lee, Peter Anderson, Devi Parikh, Dhruv Batra
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add language-only pretraining (BERT) (ablation) | VLN-BERT · R2R val unseen (path selection) | SR (%): 30.52 -> 45.17 | Improved |
| add visual grounding pretraining on web image-text pairs after language pretraining (ablation) | VLN-BERT · R2R val unseen (path selection) | SR (%): 45.17 -> 49.64 | Improved |
| add action grounding pretraining after language pretraining (ablation) | VLN-BERT · R2R val unseen (path selection) | SR (%): 45.17 -> 50.02 | Improved |
| add both visual and action grounding stages (ablation) | VLN-BERT · R2R val unseen (path selection) | SR (%): 45.17 -> 59.26 | Improved |
| add visual grounding stage to the full curriculum (ablation) | VLN-BERT · R2R val unseen (path selection) | SR (%): 50.02 -> 59.26 | Improved |
| add action grounding stage to the full curriculum (ablation) | VLN-BERT · R2R val unseen (path selection) | SR (%): 49.64 -> 59.26 | Improved |
| full pretraining curriculum vs none (ablation) | VLN-BERT · R2R val unseen (path selection) | SPL: 0.27 -> 0.55 | Improved |
| full pretraining curriculum vs none (ablation) | VLN-BERT · R2R val unseen (path selection) | NE (m): 6.81 -> 4.10 | Improved |
| add VLN-BERT to speaker + follower ensemble | speaker + follower · R2R val unseen (path selection) | SR (%): 67.90 -> 73.61 | Improved |
| add a second randomly seeded follower to ensemble | speaker + follower · R2R val unseen (path selection) | SR (%): 67.90 -> 69.14 | Improved |
| add a second speaker to ensemble | speaker + follower · R2R val unseen (path selection) | SR (%): 67.90 -> 70.58 | Improved |
| replace follower/speaker single model with VLN-BERT | single model re-ranking · R2R val unseen (path selection) | SR (%): 54.66 -> 59.26 | Improved |
Closest papers by meaning
- Airbert: In-Domain Pretraining for Vision-and-Language Navigation (ICCV 2021 · similarity 0.92)
- A New Path: Scaling Vision-and-Language Navigation With Synthetic Instructions and Imitation Learning (CVPR 2023 · similarity 0.91)
- Bridging the Visual Semantic Gap in VLN via Semantically Richer Instructions (ECCV 2022 · similarity 0.91)
- Towards Learning a Generic Agent for Vision-and-Language Navigation via Pre-Training (CVPR 2020 · similarity 0.90)
- Do Visual Imaginations Improve Vision-and-Language Navigation Agents? (CVPR 2025 · similarity 0.89)
- WebVLN: Vision-and-Language Navigation on Websites (AAAI 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
