Semantic search with frigate, indexing automation?

[restive harness]

Indexing seems to index all new images right and is a extensive task from what i read. Would there perhaps be a way to automate this task? Preferably in the night when nothing is usually going on, a indexation would be nice.

Also with multiple object detections per minute, quite common per second, how long can i expect a indexation to take place with a intel n97 cpu?

Edit: and the ram/gpu usage is only during indexing? Or also during searching?

Thanks!

[ashen cave]

You only need to index once (if at all), there's no reason to automate it

The GPU memory usage is constant as the model lives in memory but the CPU / GPU load only is there while embedding inferences are being run

And embeddings inferences are only run when the object stops being detected, not continuously, so it's not a constant load