bdaiinstitute / vlfm

The repository provides code associated with the paper VLFM: Vision-Language Frontier Maps for Zero-Shot Semantic Navigation (ICRA 2024)
http://naoki.io/portfolio/vlfm.html
MIT License
215 stars 21 forks source link

Cannot find SemExpITMPolicyV2, SemExpITMPolicyV3 #40

Closed JeffreyYH closed 2 months ago

JeffreyYH commented 3 months ago

I was trying to run the eval.py code invlfm/semexp_env, in this line (14, from vlfm.semexp_env.semexp_policy import SemExpITMPolicyV2, SemExpITMPolicyV3) of eval.py file, I was not able to find SemExpITMPolicyV2 and SemExpITMPolicyV3 in this repo. Is there any way to provide these two files and also the training files?

naokiyokoyama commented 3 months ago

Seems like it was removed in a recent commit. Here is the latest commit before it was deleted:

https://github.com/bdaiinstitute/vlfm/blob/456ccef4a0a9ba2fdc863c0371e1e015298b3e30/vlfm/semexp_env/semexp_policy.py

JeffreyYH commented 3 months ago

Thank you for your reply! Even in this commit, the two policy classes

class SemExpITMPolicyV2(SemExpMixin, ITMPolicyV2):
    pass
class SemExpITMPolicyV3(SemExpMixin, ITMPolicyV3):
    pass

are all just empty.

In addition, in the repo I could not find arguments and envs either.

from arguments import get_args 
from envs import make_vec_envs
naokiyokoyama commented 3 months ago

They are not empty, they inherit an ITMPolicy combined with a mixin class.

Those imports assume that you have the repo fort Gibson ObjectNav installed https://github.com/devendrachaplot/Object-Goal-Navigation

JeffreyYH commented 2 months ago

Thank you! So shall I use the habitat version that Object-Goal-Navigation uses? Since I encountered many errors using the version from the installation instruction of the vlfm repo.

naokiyokoyama commented 2 months ago

Yes, you have to use the habitat version they use