Closed maturk closed 10 hours ago
I can give this a try. I’ve been working with nerfstudio exclusively through docker containers for a few months now, so i can verify when I hopefully get it working. Thanks again for your help over the weekend!
I can give this a try. I’ve been working with nerfstudio exclusively through docker containers for a few months now, so i can verify when I hopefully get it working. Thanks again for your help over the weekend!
Thanks for the help @rowellz, I appreciate it.
@maturk I failed to mention earlier that I've been running nerfstudio via docker-compose
, which is my preferred method of managing and running docker containers. The official nerfstudio docs only show support for the docker run
command.
This lead me to tweaking the existing .devcontainer/docker-compose.yml
file, which hasn't been touched in over a year. I can confirm that I was able to run ns-process-data video
followed by ns-train splatfacto
after running docker-compose up
. Here is the link to the diff of my forked repo
It is going to take me some more time to resolve the versioning for the root Dockerfile, since I don't use docker run
. But if anyone is eager to get gsplat 1.0.0 running in a docker container, my forked repo linked above will support it at the moment.
I am probably going to have to wave the white flag on trying to update the root Dockerfile, my apologies @maturk. I noticed on DockerHub that nerfstudio 1.1.3 was pushed, so I tried simplifying my forked repo by referring to the dromni/nerfstudio:1.1.3
image directly in the docker-compose.yml
file. I unfortunately ran into the same issue here
The good news is the solution that @jkulhanek offered in that discussion still works, and is how I got my forked repo working with gsplat 1.0.0
in the first place. If there is any interest in adding docker-compose
to the nerfstudio docs, let me know, I'd be happy to contribute.
Hi, sorry for the delay. I plan to integrate custom docker images soon
This is what we should prioritize. I have seen multiple users saying gsplat 1.0 is incompatible with torch 2.0.
Can you help? @Zunhammer I believe we also need to update torchvision version.
pytorch==2.1.2 torchvision==0.16.2
Yeah sure, I added a PR for this.
Chaning the version is actually quite easy (see PR). Additionally, we need to take care of the setuptools version (below 70) and we need to remove the pinning of tinycudann as the version 1.6 is not compatible with the new torch version.
Let me know if you need any further assistance
Hi,
I am not a Docker user, but the torch version should be bumped up to 2.1.2 + here. Would make a PR myself but I am a Docker noob.