Research map / Vision-language and multimodal / Robot action and navigation
Multimodal Transformer with Variable-Length Memory for Vision-and-Language Navigation
Chuang Lin, Yi Jiang, Jianfei Cai, Lizhen Qu, Gholamreza Haffari, Zehuan Yuan
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 the memory-aware consistency loss to MTVM (ablation) | MTVM · R2R val unseen | SR / SPL: 64.0 -> 65.7 SR; 58.6 -> 59.4 SPL | Improved |
| use random word dropping as data augmentation instead of the consistency loss (ablation) | MTVM · R2R val unseen | SR / SPL: 64.0 -> 64.5 SR; 58.6 -> 57.8 SPL | Mixed |
| replace the recurrent state history with the cross-modal activation memory bank | VLN-BERT · R2R val unseen | SR / SPL: 63 -> 65.7 SR; 57 -> 59.4 SPL | Improved |
| use cross-modal memory instead of hierarchical visual-only history (HAMT) | HAMT · R2R val unseen | SR / SPL: 63.5 -> 65.7 SR; 57.5 -> 59.4 SPL | Improved |
| use memory with single-direction cross-attention (MTVM-dagger*) instead of VLN-BERT's hidden-state network | VLN-BERT · R2R val unseen | SR / SPL / memory: 63.3 -> 63.6 SR; 57.5 -> 58.2 SPL; 8.6GB -> 8.4GB | Improved |
| use bidirectional cross-attention instead of language-only keys/values in MTVM (ablation) | MTVM without consistency loss · R2R val unseen | SR / SPL / memory: 63.6 -> 64.0 SR; 58.2 -> 58.6 SPL; 8.4GB -> 17.9GB | Mixed |
| increase the memory bank size / use variable-length memory storing every step (ablation) | MTVM · R2R val seen and unseen | NE, SR, SPL | Improved |
| raise the word dropping rate from 0.1 to 0.5 (ablation) | MTVM · R2R val seen and unseen | SR, SPL | Improved |
| raise the word dropping rate from 0.5 to 0.7 (ablation) | MTVM · R2R val seen and unseen | SR, SPL | Hurt |
| vary consistency loss weights lambda_s and lambda_m in the range 2 to 8 (ablation) | MTVM · R2R val unseen | SPL | No clear effect |
| use cross-modal history (VLN-BERT recurrent state) instead of visual-only oriented observations (E.T.) | history encoding · R2R val unseen | SR / SPL: 59.0 -> 63 SR; 54.5 -> 57 SPL | Improved |
Closest papers by meaning
- History Aware Multimodal Transformer for Vision-and-Language Navigation (NeurIPS 2021 · similarity 0.91)
- VLN BERT: A Recurrent Vision-and-Language BERT for Navigation (CVPR 2021 · similarity 0.90)
- Towards Learning a Generic Agent for Vision-and-Language Navigation via Pre-Training (CVPR 2020 · similarity 0.90)
- Episodic Transformer for Vision-and-Language Navigation (ICCV 2021 · similarity 0.90)
- Towards Long-Horizon Vision-Language Navigation: Platform, Benchmark and Method (CVPR 2025 · similarity 0.89)
- Vision-Dialog Navigation by Exploring Cross-Modal Memory (CVPR 2020 · similarity 0.89)
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.
