AI Paper Atlas

Research map / Vision-language and multimodal / Robot action and navigation

Airbert: In-Domain Pretraining for Vision-and-Language Navigation

Pierre-Louis Guhur, Makarand Tapaswi, Shizhe Chen, Ivan Laptev, Cordelia Schmid

ICCV 2021 · Vision-language and multimodal › Robot action and navigation · approach: Reasoning-enhanced vision-language navigation · 158 citations (OpenAlex)

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 changedWhereMeasuredEffect
add shuffling loss during BnB pretraining (ablation)Airbert pretraining · R2R val unseenSR: 64.21 -> 66.52Improved
add shuffling loss during R2R fine-tuning (ablation)Airbert · R2R val unseenSR: 66.52 -> 66.90Improved
add shuffling loss when pretraining with BnB and Speaker data (ablation)Airbert · R2R val unseenSR: 65.52 -> 68.67Improved
insert captionless images at random locations in BnB paths (ablation)Airbert pretraining · R2R val unseenSR: 62.45 -> 66.05Improved
add instruction rephrasing, image merging and captionless insertion over concatenation (ablation)Airbert pretraining · R2R val unseenSR: 62.71 -> 66.52Improved
use naive concatenation instead of raw image-caption pairs (ablation)Airbert pretraining · R2R val unseenSR: 62.45 -> 62.71No clear effect
use instruction generation instead of rephrasing (ablation)Airbert pretraining · R2R val unseenSR: 63.35 -> 63.11Hurt
add wrong trajectories as extra negatives in fine-tuning (ablation)VLN-BERT · R2R val unseenSR: 59.26 -> 59.11No clear effect
add keyword highlighting loss in fine-tuning (ablation)VLN-BERT · R2R val unseenSR: 59.26 -> 61.37Improved
add hard negatives from masked-keyword replacement (ablation)VLN-BERT · R2R val unseenSR: 59.26 -> 61.63Improved
add shuffled visual paths as negatives in fine-tuning (ablation)VLN-BERT · R2R val unseenSR: 59.26 -> 61.98Improved
replace VLN-BERT with Airbert pretraining in ensemble with speaker and followerAirbert ensemble · R2R val unseenSR: 73.61 -> 75.01Improved

Later papers that compared with this method

5 later accepted papers report Airbert as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy