Code for VLN-CLASH

---

Note: This codebase is currently under development. As the associated paper is still under review, we are in the process of cleaning and releasing the code progressively. Thank you for your understanding and interest.

---

1. Fine-tuning

To fine-tune the model, run:

bash run_r2r/main.bash train

---

2. Validation

To evaluate the model on the validation splits, run:

bash run_r2r/main.bash eval val_seen_unseen 

This command evaluates both the val-seen and val-unseen splits simultaneously and records the results separately.

---

3. Inference

To generate test predictions, run:

bash run_r2r/main.bash infer

This will generate the pred.json file for the test-unseen split, which can be submitted to the online VLN-CE leaderboard for fair comparison.