Hi everyone,
I am currently working on a project idea. The goal is to build a 3D modeling (photogrammetry) pipeline that scales easily and runs inside containers.
The main idea is to create a Linux-based Docker image with the RealityScan/RealityCapture CLI. This setup will run on servers with powerful NVIDIA graphics cards. Users should be able to spin up a container, drop in a folder of photos, and automatically get a 3D model out. Whether this is triggered by a script or manually doesn’t matter right now—the priority is a stable backend.
My experience so far:
I’ve been testing this locally using WSL2, both directly in the system and inside Docker containers. However, it is not running smoothly at all. Vulkan and CUDA keep acting up. The connection between the virtual Linux layer and the actual graphics card keeps breaking. The CLI often crashes or freezes in the background. This happens because the software tries to open a login screen or a pop-up window, which obviously fails in a headless Docker container without a screen. But also the silent mode and other options did not help.
My questions to the community:
1. Is a Linux/Docker setup actually a reliable path forward for production servers? Or are workarounds like Wine (to run Windows apps on Linux) just too buggy and patched together for this kind of env.?
2. Even if it is hard to set up, is this somehow possible or are there any barriers that just prevent it from working?
3. Or should I just drop the Linux plan entirely and focus on automating the pipeline directly on some kind of Windows solution?