Observed VRAM Memory Leak

[xD0135]

🔘 Operating system

Linux (Arch-like)

🔘 Runtime enviroment

PyAPP Executables (experimental)

🔘 Python version

N/A

🔘 GPU Vendor

NVIDIA Desktop

🔘 Description

Version:
$ ./depthflow-linux-amd64-v0.8.0.bin --version
DepthFlow 0.8.0 linux-amd64

I observed a memory leak when continuously rendering the video after playing around with the animation settings.

This is when I simply click the render button for the first time, everything looks good:

This is after the 4th click on Render

And this is after the 8th click on Render

To reproduce this issue, simply keep clicking on the Render button and monitor your VRAM usage, and you'll see the issue.

🔘 Traceback

No response

[Tremeschin]

Oh yea, I did some recent work on fixing a couple dumb memory leaks here and there, the one you're describing (vram) is probably fixed on this commit BrokenSource/ShaderFlow@6080f4b not present on the v0.8, as well as (system ram) on BrokenSource/ShaderFlow@ca6c33e

I'm a bit of far away from a v0.9 release as there's lots of changes and additions, you could run directly from source https://brokensrc.dev/get/source/ for fixes and new features https://brokensrc.dev/about/changelog/#0.9.0-depthflow temporarily, beware dragons! (I'll be on a month long road trip on dec 20th, could be a 2025 release 😅)

I'm working internally on properly sharing models, thread safety, etc this week. I should add a jobs queue to the webui to avoid recreating the scenes and for thread workers to render the jobs, will be much more leak-resistant, no recreation of opengl contexts, faster render times (no init), etc.

[xD0135]

Awesome! Thank you so much for this great work! Feel free to close, or keep it open until v0.9.0 is released for tracking purposes 😺

[Tremeschin]

I'll keep it open and move to depthflow repo (even tho it's mainly the other proj issues) for awareness until the 0.9 yea :)