real-stanford / cow

[CVPR 2023] CoWs on Pasture: Baselines and Benchmarks for Language-Driven Zero-Shot Object Navigation
https://arxiv.org/abs/2203.10421
108 stars 7 forks source link

Training Learnable Exploration Baselines #13

Open kellyiss opened 1 year ago

kellyiss commented 1 year ago

Thanks for your helpful work! I see that there is an implementation of clip-based actor critic exploration model https://github.com/real-stanford/cow/blob/main/src/models/exploration/clip_ac.py

Where can I find the training code for the learnable exploration baselines?

sagadre commented 1 year ago

thanks for the issue, and sorry for the delay🙂 have been a bit pre-occupied with the endless stream of conferences and deadlines (most recently ICCV travel)

have blocked some cycles to look into this this week. fundamentally, something like the following should work:

allenact src/allenact_experiments/runable/exploration_robothor_rgb_vitb32clipgru_supervised.py
johncruyff14 commented 1 year ago

Hi Kelly and Samir @sagadre . Do you resolve this issue? https://github.com/real-stanford/cow/issues/4 If yes, do you mind sharing the solution to us?😊. I am asking because I saw lots of people get stuck and we really get stuck for a long time.