CompletionFormer icon indicating copy to clipboard operation
CompletionFormer copied to clipboard

Some questions about the implementation of this repository with cuda 12.9

Open Tony-syr opened this issue 3 months ago • 1 comments

I have an ubuntu 24.04 which support cuda 12.9, I can not use cuda 11.3 Have you tested the code for cuda 12.9? How to prepare the environment with cuda 12.9?

Tony-syr avatar Sep 22 '25 08:09 Tony-syr

Hi, do you have any progress now? I have implemented this repo with pytorch 2.3.1+ cuda 12.1. I have only tested on NYUv2 dataset till now. The results are similar to those in the paper when the number of sampling points is 500, but with 200 / 50 points, the RMSE rose quickly. I replaced DCNv2 module using code from https://github.com/lbin/DCNv2. Not sure is pytorch or DCNv2 affects the performance.

ShirleyQSY avatar Nov 27 '25 05:11 ShirleyQSY