OpenFly
A large-scale dataset and navigator framework for Aerial Vision-and-Language Navigation (Aerial VLN). Its training data includes finely annotated flight trajectories aligned with natural language instructions.
Timeline (last 90 days)
On the frozen OpenFly navigator, when input was switched from the exhaustive commands used in training to user-style brief instructions, the success rate dropped from 31.03% to 11.33%.
OpenFly is a currently large-scale aerial VLN dataset and navigator framework whose training instructions are typically written by annotators, containing details such as direction, distance, and landmarks, and are aligned segment by segment with flight trajectories.
Using only synthetic weak instructions fed directly into the navigator, the success rate was 15.27%, falling between the exhaustive command rate of 31.03% and the real user instruction rate of 11.33%.
On an unseen OpenFly test set, TGIT improved the success rate from 4.95% to 20.79%.
All Facts (4)
On the frozen OpenFly navigator, when input was switched from the exhaustive commands used in training to user-style brief instructions, the success rate dropped from 31.03% to 11.33%.
50%UnverifiedOpenFly is a currently large-scale aerial VLN dataset and navigator framework whose training instructions are typically written by annotators, containing details such as direction, distance, and landmarks, and are aligned segment by segment with flight trajectories.
50%UnverifiedUsing only synthetic weak instructions fed directly into the navigator, the success rate was 15.27%, falling between the exhaustive command rate of 31.03% and the real user instruction rate of 11.33%.
50%UnverifiedOn an unseen OpenFly test set, TGIT improved the success rate from 4.95% to 20.79%.
50%