search

nickgkan / 3d_diffuser_actor

Code for the paper "3D Diffuser Actor: Policy Diffusion with 3D Scene Representations"
https://3d-diffuser-actor.github.io/
MIT License
159 stars 16 forks source link

Training cost on the RLBench Dataset. #22

Closed siaoliu closed 2 months ago

siaoliu commented 2 months ago

Hi, Thanks for your wonderful work! I would like to inquire about the training cost on the RLBench dataset. Does the model have significant efficiency advantages compared to the PerAct.

twke18 commented 2 months ago

Hi,

Yes, our model is much more efficient than PerAct. Using the same setup as PerAct, it takes us 6.5 days for training our model, while, PerAct needs a training time of 16 days. Currently, our code base is bottlenecked by IO, which need some engineering effort to improve.