[BUG] Cellpose hogging the CPU during multi-threading

[BennyGinger]

Describe the bug I integrated cellpose in my analysis pipeline, which mainly preprocess images then use cellpose to segment. During the preprocess I use multithreading a lot (mainly to extract image from large stack). I recently upgraded my conda env with the newest cellpose 3.0, since then only 2 CPU are used at a time (compared to 16 with cellpose 2.0). I can see in that all processes are sent (many python processes appear when I launch the script) but all are inactive but 2.

I can isolate my image preprocess module and it works fine (with 16 CPU). Just by importing cellpose reduce the CPU usage to 2.

Any suggestions?