cshizhe / HM3DAutoVLN

Official implementation of Learning from Unlabeled 3D Environments for Vision-and-Language Navigation (ECCV'22).
34 stars 1 forks source link

KeyError '64_2' #4

Open Feliciaxyao opened 1 year ago

Feliciaxyao commented 1 year ago

Thank for the great work! When I try to follow the instruction in README.md to finetune the pretrained model, the error always occurs and prevents the code from running properly.

Listener training starts, start iteration: 0 Progress: |██████████████████████████████████████████████████| 99.8% Complete total_actions 1, max_length 1, entropy 0.0000, IL_loss 4.4567, OG_loss 1.1653, policy_loss 0.0000, critic_loss 0.0000 eval 188 predictions eval 123 predictions eval 1423 predictions eval 3521 predictions eval 6292 predictions Traceback (most recent call last): File "main_nav_obj_hm3d.py", line 310, in <module> main() File "main_nav_obj_hm3d.py", line 304, in main train(args, train_env, val_envs, aug_env=aug_env, rank=rank) File "main_nav_obj_hm3d.py", line 214, in train scoresummary, = env.eval_metrics(preds) File "/root/mount/Matterport3DSimulator/VLN-DUET/map_nav_src/reverie_hm3d/env.py", line 407, in eval_metrics scan, gt_traj, gt_objid = self.gt_trajs[instr_id] KeyError: '64_2'

It seems like some information is missing. Have you ever been in this situation, or do you have some advice on how to solve this issue?