search

VVRPanda / ExpPoint-MAE

10 stars 0 forks source link

MoCo module #1

Open JiayuLiu666 opened 1 year ago

JiayuLiu666 commented 1 year ago

Hello, have you included moco training modules in the code? I did not find the moco experiment.

ColonelVLS commented 1 year ago

Hello, have you included moco training modules in the code? I did not find the moco experiment.

Hey, we are actively refactoring the code and will upload each piece when it is in "viewable" condition. Check back next week and it should be up!

JohnRomanelis commented 1 year ago

Hello! Sorry for the delayed responce. We were really busy lately with some other running projects and we didn't have the time to properly fomat our codebase. Not to keep you waiting any longer, I am attaching to this message the code used to train the hybrid model (I am using a .txt file instead of a .py because it does not allow me to upload the latter in a responce).

In this code you just have to set the weight for the MAE loss to zero and you get the MoCo variant. It is a bit slower to run, as you also run the MAE pipeline. If you want, you can try to comment out the MAE related code. I believe the comments in the code will help you do this.

I am also attaching the required configuration file. Please note that this version of the cfg file may not be the one used to produce our paper results. I haven't had the time to check it.

MoCo&MAE.txt config_file.txt