Uni-NaVid for SatNav — Navigation-pretrained initialization
This checkpoint is designed for use with SatNav, the continuous-state vision-and-language navigation platform for satellite maps. The SatNav Uni-NaVid integration uses it to predict short action sequences from a language instruction and a stream of RGB observations.
Model
- Starting checkpoint:
Jzzhang/Uni-NaVid - Vision encoder: EVA-CLIP
- Training data: SatNav-v0.1 offline trajectories
- Training: 1 epoch, learning rate
1e-5 - Action window: up to 4 actions
Repository name
uninavid-satnav: Uni-NaVid adapted to SatNavcontinue: navigation-pretrained initialization from the Uni-NaVid checkpoint1ep: trained for one epochlr1e-5: uses a learning rate of1e-5
Reference Results
Results reported in the SatNav paper, baseline comparison, row UniNaVid (navigation-pretrained).
SR, SPL, and OS are percentages. NE is measured in meters. Steps is the average number of executed environment actions, including turns and stop.
| Split | SR (%) | SPL (%) | OS (%) | NE (m) | Steps |
|---|---|---|---|---|---|
Test Seen (val_seen) | 49.7 | 49.1 | 68.2 | 87.11 | 53.99 |
Test Unseen (val_unseen) | 36.7 | 36.3 | 55.9 | 149.85 | 64.75 |
Usage
Use this checkpoint with the SatNav Uni-NaVid integration. It is also listed in the SatNav Baseline Model Zoo.
License
The checkpoint follows the applicable Llama 2 terms. Uni-NaVid, EVA-CLIP, SatNav, and other components retain their own licenses.