FR: Parallelized model optimization in installer

The model optimization in the Zed SDK installer takes forever, but I noticed its only using a single CPU. I have 11 others in this machine sitting idle.

Would it be possible to change the installer so that it optimizes more than one model at a time, to take advantage of multiple cores? Or is the GPU the limiting factor here?

Thanks!

Hi,

Running multiple model optimizations in parallel is not recommended.
Indeed, the optimization’s goal is to find the fastest implementation of a given AI model architecture. Having multiple optimization processes simultaneously will increase the GPU load and provide a non-optimal result, decreasing the overall AI model performance.

We understand the process can take a long time to fully run, but it must be done once per model, at the ZED SDK installation or its first use.

Best regards,
Benjamin V.

Stereolabs Support