Does anyone know if the RealityScan frame extraction from video algorithm only skips non-intra frames or, has a routine for detecting motion blur or other defects within an individual frame (thus making them unsuitable) been added of late? Trying to avoid manual assessment/discarding of dud frames if possible. Thanks in advance.
Hi @DavoQUT , RealityScan doesn’t check frames for motion blur or other defects when importing a video. It takes frames at a fixed time interval, which you set with Jumps length in the import dialog (or the jumpsLength parameter of the importVideo CLI command). Every frame it picks goes into the project, sharp or not.
What I’d suggest for now is a short Jumps length, so you get more frames than you need. Then remove the blurry ones before alignment, either by hand or with an external tool that scores frame sharpness. You can also extract the frames with that tool and import them into RS as regular images.
I can’t make any promises, but I’ll pass this on to the devs as a feature request.