Open ZuoJiaxing opened 4 years ago
The normalization guarantees that the location is related to the current "front" view. In the paper of Speaker-Follower Models for Vision-and-Language Navigation, the "front" view is always set with elevation=0. We follow the practice here.
https://github.com/airsplay/R2R-EnvDrop/blob/c416108e1f713a5fea2e2bda887ab22080fb3cd1/r2r_src/env.py#L265
I'm curious about why the elevation of the navigable candidates is not updated as the heading, such as: loc_heading = normalized_heading - base_heading loc_elevation = normalized_elevation - base_elevation