Research map / Vision-language and multimodal / Robot action and navigation
Harnessing Input-Adaptive Inference for Efficient VLN
Dongwoo Kang, Akhil Perincherry, Zachary Coalson, Aiden Gabriel, Stefan Lee, Sanghyun Hong
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 |
|---|---|---|---|
| apply existing early-exit method MuE to the HAMT ViT | HAMT · R2R val unseen | SR: 66.16 -> 43.93 | Hurt |
| apply MuE | HAMT · R2R val unseen | GFLOPs: 4763.24 -> 4409.62 | No clear effect |
| add k-extension to process only navigable views plus k neighbors (ablation) | HAMT · REVERIE val unseen | GFLOPs (+C): 5434.71 -> 3121.20 (SR 31.81 -> 24.96) | Improved |
| add LSH caching on top of k-extension (ablation) | HAMT · REVERIE val unseen | GFLOPs (+C): 3121.20 -> 2359.72 | Improved |
| add adaptive thresholds on top of k-extension (ablation) | HAMT · REVERIE val unseen | GFLOPs (+C): 3121.20 -> 2723.01 (SR 24.96 -> 24.60) | Improved |
| increase k from 1 to 6 (ablation) | k-extension · R2R val unseen | SR: 54.32 -> 62.84 (GFLOPs 1250.65 -> 2414.46) | Improved |
| increase early-exit aggressiveness A from 0 to 0.022 (ablation) | adaptive thresholding · R2R val unseen | SR: 61.17 -> 55.60 (GFLOPs 2013.48 -> 1768.85) | Hurt |
| swap cosine similarity for SSIM, FSIM, LPIPS or ViT activation in LSH (ablation) | LSH · R2R val unseen | SR: 60.41 -> 60.49 (LPIPS) | No clear effect |
| add LSH caching to scan-only SGM in continuous VLN (ablation) | VLN-CE BERT · R2R-CE val unseen | SR: 39.86 -> 37.14 (GFLOPs 2396.54 -> 1741.33) | Hurt |
| apply scan-only subgoal generation module to predict navigable views | VLN-CE BERT · R2R-CE val unseen | GFLOPs: 18074.05 -> 2396.54 | Improved |
| apply visual corruption (Speckle Noise) to efficient agent | efficient HAMT · R2R val unseen | SR: 60.41 -> 40.87 | Hurt |
Closest papers by meaning
- Towards Open Environments and Instructions: General Vision-Language Navigation via Fast-Slow Interactive Reasoning (CVPR 2026 · similarity 0.91)
- DecoVLN: Decoupling Observation, Reasoning, and Correction for Vision-and-Language Navigation (CVPR 2026 · similarity 0.90)
- Toward a Causally-Inspired Evolving World Model for Vision-and-Language Navigation in Continuous Environments (TPAMI 2026 · similarity 0.90)
- General Scene Adaptation for Vision-and-Language Navigation (ICLR 2025 · similarity 0.90)
- Test-Time Adaptation for Online Vision-Language Navigation with Feedback-based Reinforcement Learning (ICML 2025 · similarity 0.89)
- Fast-Slow Test-Time Adaptation for Online Vision-and-Language Navigation (ICML 2024 · 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.
