Feedback wanted: Should preflight be a command instead of always on?

I’m thinking of ways to speed up preflighting, especially for folks who have 2k+ images. And I’m wondering if maybe preflight should be something that’s done on command instead of an always on thing?

The advantage of this is that the CPU would no longer burn cycles just loading up images when they are going to have to be thrown out in a few seconds anyway, since you’re messing around with nodes.

Does anyone have an opinion one way or another? Do you like the always on preflight or not?

For me pre-flight should be an option.

I’d also go with option.

Maybe right next to the info-area at the top as a button that can quickly be toggled on and off. If it was in the preferences it would be more effort if you just wanted to en- or disable it briefly (I could image this to be a fairly common usecase).

Yes, please. Yes. I regularly process upwards of 90k batches. I have to split them into folders and then open ~10 .retrobatch projects at the same time, run them in tandem, and hope for the best. This is only slow when using classification, which is exactly what I’m doing – processing classification for ML training. I train smaller models for filtering and do multiple passes until I get the very best (EfficientNet and MobileNet benchmark on speed the same, but EN is far more accurate). Otherwise RB would just chug through it mind-bogglingly crazy fast as usual. I’m on a M1 Max 64GB on the latest RB Beta and preflight will make it swap another 50 or 60 extra GB under these conditions. Cheers :metal: