[EN] Issues with ImageOps - VRAM and concurrency

[JeS24]

Feature request / Suggestion

Currently, we have set celery's concurrency to 1 process with several threads, in order to get around race conditions and OutOfMemory issues with the deep learning models used for ImageOps. The problem here is that ImageOps is too slow when run serially. Figure out a way to parallelize image checks, while balancing RAM / VRAM usage.

Possible implementation

Chunking might help here, the same way it works with MeshOps. Note that subprocess would not work with celery most likely, and threading on GPU has its own share of issues.

Self-check

• I have checked smlab-niser/tirtha-public and could not find any related open or closed feature request.

[github-actions]

This issue is stale because it has been open for 60 days with no activity. Remove stale label or comment to re-open.

[github-actions]

This issue is stale because it has been open for 60 days with no activity. Remove stale label or comment to re-open.